eBook

5G Filtering Solutions

September 2019

S P O N S O R E D B Y
Table of Contents

3

Introduction
Pat Hindle
Microwave Journal, Editor

4

Approaching the 5G mmWave Filter Challenge
Peter Matthews
Knowles Precision Devices

Designing a Narrowband 28 GHz Bandpass Filter for
5G Applications
David Vye and John Dunn, NI AWR Group • Dan Swanson, DGS Associates
Jim Assurian and Ray Hashemi, Reactel Inc. • Philip Jobson, Design Consultant

15

Tolerance and Size Analysis for mmWave Filter
Manufacturing
Knowles Precision Devices

21

Global 5G Rush But No Global 5G Handsets
Ben Thomas
Qorvo, Greensboro, N.C.

25

SAW/BAW New Market Entrants Offer New Approaches
With Contributions from: Akoustis Technologies, OnScale, Resonant

32

Reduce Cost and Complexity in 5G mmWave systems with
Surface Mount Solderable Filter Components
Knowles Precision Devices

2
Introduction

5G Filtering Solutions

4G LTE added many new frequency bands that were closely spaced with other communications
bands and forced filter innovation in the SAW and BAW area for more stable and selective filters.
These technologies perform well in the lower GHz frequency ranges but are severely challenged at
wider bandwidths and higher frequencies. Newer BAW technologies such as FBAR and some other
new substrate technologies are being developed that can handle wider bandwidths and frequencies
up to about 6 GHz but cannot function in the mmWave range. This ebook takes a look at the filtering
challenges and design tradeoffs for 5G plus some of the newer technologies being developed to handle
the wider bandwidths and higher frequencies that will be required for 5G.

The first article takes a good look at the challenges and design tradeoffs needed to address the
5G market including solutions that address the mmWave frequencies. It covers various surface mount
microstrip filters and their capabilities to address high frequencies. The next article steps through
the design of a 28 GHz bandpass filter for 5G including simulation and port tuning techniques. It
also presents measured results of a manufactured proto-type filter. The next article is a white paper
addressing tolerance and size analysis for mmWave filter manufacturing to give you an idea of the
importance of dimensional requirements.

The next article is on a little different subject as it discusses the challenges to designing a global
5G handset due to the large number of frequency bands worldwide and the variety of bands used by
each country. Then we have an article that looks at several new entrants into the SAW/BAW market
with companies taking new approaches to filtering. Some are using new materials/configurations while
others are using new simulation techniques to achieve better filter design in less time. The last article is
a white paper about reducing the cost and complexity in 5G mmWave systems with surface mount filter
components.

We hope that this ebook provides a broad look at the filtering challenges and solutions that address
the wider bandwidths and higher frequencies required by 5G. We would like to thank Knowles Precision
Devices for sponsoring this ebook so that is it free to download by anyone interested in reading about
these technologies.

Pat Hindle, Microwave Journal Editor

3
Approaching the 5G mmWave
Filter Challenge
Key specifications for mmWave filtering and available options
Peter Matthews
Knowles Precision Devices

I
n the world of LTE, developers are very familiar The focus has now shifted to solving the practical is-
with the available filtering technologies that work, sue of how to build an actual mmWave-capable base
namely surface acoustic wave (SAW) and bulk station and implement high performance RF filtering.
acoustic wave (BAW) filters. These filters cover a Fortunately, mmWave technology has been employed
range of frequencies up to 6 GHz, come in small sizes for decades in various fields and functions. For example,
and offer good performance-to-cost trade-offs, making it has a long history in military, aerospace and SATCOM
them the dominant off-chip approaches in mobile de- applications such as K-Band inter-satellite communica-
vices today. Unfortunately, analogous filtering options tion and ranging and Ka-Band high-resolution radar.
for the mmWave spectrum have issues regarding vi- In other industries, the automotive field is using both
ability, performance, size and availability, while research 24 GHz in short-range and 77 GHz in long-range radar-
teams helping to write 5G standards have yet to pro- based advanced driver assistance systems (ADAS) to scan
vide information on what filters will be required, where a vehicle’s environment for driver support or automated
they need to be placed in the base station and what driving. In the U.S., 36 to 40 GHz is currently licensed for
performance metrics they must meet. high speed microwave data links between a cellular base
station and a base station controller; and, the unlicensed
CONSIDERATIONS 60 GHz band is used in short range data connections and
The obstacles to and advantages for using the IEEE 802.11ad WiGig for high bandwidth streaming.
mmWave spectrum are both widely-publicized and
well-understood. High frequencies suffer from range FILTER REQUIREMENTS
limitations and path loss through air, objects and build- Chief concerns of industry leaders include:
ings. However, mmWave signals require much smaller • Quality and Performance: Can the filter perform ac-
antennas, which can be tightly packed together to cre- curately, repeatably and reliably across thousands of
ate single, narrowly focused beams for point-to-point units?
communication with greater reach. Frequency bands • Time-to-Market: 5G deployment is nearly at hand,
around 28, 38 and 72 GHz are the main candidates for so do we have a solution that is available today for
5G mmWave, having demonstrated directional anten- rapid prototyping and product development?
na, beamforming and beam tracking performance in • Ease of Integration: Will the new filter solution be
multipath environments.1-2 relatively straightforward to implement into existing

www.mwjournal.com/articles/32228
4
 technology? Can it be easily adapted for use with systems leads to densely populated boards, where
various wireless standards and frequencies? heat from surrounding components can affect filter
Specific filter performance metrics must address the stability.
following: • Power Handling: The ability of the filter to with-
• Percent Bandwidth: The filter technology must not stand large amounts of transmit RF power is mostly
limit the radio access system bandwidth. a concern in traditional macro-cell use cases below 3
• Selectivity: High selectivity enables designers to GHz. In mmWave 5G use cases, the transmit power
make good use of available bandwidth. is spread over the individual array elements and the
transmit range is reduced as well.
• Insertion Loss: Power is a system cost driver and on
the receiver side, insertion loss impacts the overall • Filter Placement: As shown in Figure 2, there are
noise figure of the receiver. several practical locations: at the antenna element
(Position 1), behind the amplifiers closest to the an-
• Size and Packaging: In phased arrays, the antenna
tenna (Position 2) and on the high frequency side of
elements must be sufficiently close together to avoid
the mixers (Position 3). It should be noted that the
generating grating lobes; a half wavelength spacing
dashed lines represent potential 1:N branching in a
for mmWave frequencies amounts to only a few mil-
beam forming architecture, and that Figure 2 shows a
limeters. As shown in Figure 1, phased arrays often
simplified single path from mixer to antenna. The ad-
used in mmWave applications use a plank architec-
vantage of the Position 1 implementation is that lin-
ture, in which the gold areas indicate the antennas
earity and bandwidth design constraints for the sub-
mounted on a printed circuit board (PCB) (green
sequent blocks is eased, potentially reducing overall
area) and the blue areas indicate the circuit “planks”
system cost. Filtering placed at Position 1 suppresses
extending 90 degrees from array (where space for
noise far from the channel of interest across a wide
filters is already very limited). Base station manufac-
frequency range, making wideband suppression and
turers, however, are now looking into flat panel archi-
very low insertion loss key performance features.
tectures where the circuitry is implemented on the
Given that each sub-array has its own filter, small size
back side of the antenna PCB, requiring even denser
and low cost are important.
spacing for filtering and other functional blocks.
• Temperature Stability: In order to make efficient use SOLUTIONS
of available bandwidth, the filter must meet its speci- SAW and BAW filters have long dominated the off-
fications over a range of temperatures. Small-scale chip filtering market in mobile devices because of their
systems may be deployed in exposed environments excellent performance specs, tiny footprints and afford-
that experience extremes in temperature and tem- able prices compared to other options. Unfortunately,
perature variation. Further, overall size reduction in the nature of their design—using interdigital transduc-
ers (IDT) to process signals as acoustic waves—exhibits
degradation in selectivity at frequencies greater than 6
GHz, making SAW and BAW technology unfeasible for
mmWave applications. Nevertheless, it is worth noting
their performance metrics (see Table 1) as a benchmark
for potential mmWave filter solutions.

Mixer RF Filter Switch

3 2

Splitter
Mixer RF Filter LNA Combiner s Fig. 2 Potential radio filter locations.

TABLE 1
TYPICAL SAW AND BAW FILTER PERFORMANCE
(a)
Typical SAW/BAW
Performance Goals
Performance
Low Insertion Loss < 3 dB
Excellent Rejection > 30 dB out of band
Broad Bandwidth Up to 100 MHz
(b)
Small Size ~9 mm2
s Fig. 1 Alternative array architectures: plank (a) and flat Good Temperature Stability ~3 ppm/ºC
panel (b).
5
 Proven Approaches
Figure 3 summarizes the frequency ranges of today’s
commonly available filters. At higher mmWave frequen-
cies, acoustic filtering is not as practical, so many de-
SAW (to 2.5 GHz) velopers have turned to electromagnetic (EM) solutions.
BAW (to 6 GHz) For mmWave applications at 20 GHz and higher there
FBAR (to 10 GHz) are dielectric and cavity waveguide, on-chip and mi-
Dielectric Waveguide (to ∼30 GHz) crostrip (or planar thin film) filters.
Waveguide (to ∼80 GHz)
Waveguide filters are hollow or dielectric filled cavi-
ties within metal tubes that act as BPFs, blocking certain
On-Chip (to ∼70 GHz)
wavelengths while allowing others to pass. Character-
Microstrip (to ∼70 GHz) ized by high-power handling and low loss, waveguide
Frequency
filters are widely used for mmWave frequencies from 20
to 80 GHz in military, radar, satellite and broadcasting
s Fig. 3 Frequency ranges of common filter technologies. markets. Unfortunately, waveguides typically have di-
mensions in the centimeter range (see Figure 4). A λ/2
Emerging SAW/BAW Alternatives array element spacing at 28 GHz in free space is 5.35
Given the successful use of SAW and BAW technolo- mm. Until manufacturers are able to sufficiently reduce
gies at lower frequencies, researchers are naturally look- waveguide sizes while still meeting electrical perfor-
ing into developing an acoustic equivalent for mmWave mance requirements, this solution may not be practical
applications. for an array antenna system.
The Film bulk acoustic resonator (FBAR) filter is a On-chip filters using semiconductor technologies are
form of BAW filter that can reportedly operate from 5 attractive because of the compact circuits, tight toleranc-
to 20 GHz,3 which is applicable for LTE but still below es, and the capability for integration with other devices
desired mmWave ranges. Resonant Inc. is developing a to form system on a chip (SoC) solutions. Yet, significant
so called XBAR™ technology that seeks to outperform performance issues regarding Q-factor, losses and noise
FBARs, but currently it is only available as licensable in- figure (NF) occur with the production of on-chip devices
tellectual property for manufacturers to complete devel- having reduced dimensions for mmWave frequencies.
opment. Challenges arise from various factors, including the
The use of substrate integrated waveguides (SIW) of- physical characteristics of the semiconductor material
fers another approach where researchers seek to create and the cost of implementation. For example, GaN cir-
small waveguide cavity filters for use in PCB and on-chip cuits are made as thin as possible to increase heat dis-
applications. While attractive for its wideband properties, sipation; however, filter Q is proportional to dielectric
good isolation and lower losses, challenges for mmWave substrate thickness, so the high-power advantage of
include radiation leakage between plated through-holes using GaN devices works in opposition to integrating
(PTH), difficulties designing SIW transitions and dimen- filters with high Q. In addition, a filter structure in GaN
sional variations in the PTH side walls (which are detect- occupies valuable real estate on the wafer that would be
able at mmWave frequencies). 4
better allocated to active devices. Building on-chip high
Micro-Electro-Mechanical Systems (MEMS) offers the
Q filter structures to serve in a front-end application is
potential for tiny, tunable RF filters created using con-
currently impractical.
ventional semiconductor fabrication processes. The
methodology is interesting due to its dimensional accu- Microstrip filters have been considered for mmWave
racy, high component density and low-cost at high vol- applications but are commonly dismissed for various
5
umes; however, current designs are still in the research performance issues. Note that there are at least three
phase or are limited to the sub-mmWave ranges. different form factors:
• Microstrip on PCB
• Microstrip in a multilayer, low tem-
perature co-fired ceramic (LTCC)
package
28 GHz Waveguide: ∼1450 mm2
• Microstrip in a small form factor, sin-
gle-layer package
28 GHz Microstrip on PCB (Rogers 4350): ∼148 mm2
Microstrip filters printed on PCB are
Knowles 28 GHz Microstrip on Alumina: ∼55 mm2 appealing because of their simple con-
Knowles 28 GHz Microstrip on PG: ∼43 mm2 struction, but high performance PCB
Knowles 28 GHz Microstrip on CF: ∼22 mm 2 Smartphone solutions generally reach centimeter-
10,000 mm2 range sizes, which are much larger
Knowles 28 GHz Microstrip on CG: ∼10 mm 2

1.8 GHz SAW: ∼9 mm 2 than the sub-wavelength dimensions

required for mmWave antennas. Varia-
tions caused by the PCB manufactur-
ing processes also limit performance,
s Fig. 4 Sizes of mmWave filters compared to typical SAW filters.

6
resulting in higher insertion loss and lower suppression sion lines to create high performance resonant struc-
values. tures. With a careful choice of filter topology and
Another option is surface mount technology (SMT). dielectric materials, high rejection, low loss filters that
SMT assembly has been long used in commercial sys- are temperature stable from −55°C to 125°C can be
tems and is now being adopted in mmWave military produced. These filters provide performance similar
systems for potential cost savings. Unlike chip and wire to their lower frequency SAW and BAW counterparts
solutions, SMT filters have standardized form factors to (see Table 2). High performance and small form factor
reduce overall assembly time, and require no post-tun- devices are possible within the constraints of 5G New
ing. Figure 5 shows the performance repeatability of Radio (NR) systems. Figure 6 shows the functional-
100 SMT BPFs, measured in the 26.5 to 29.5 GHz range ity of a low loss 26 GHz filter with greater than 50 dB
without tuning. of suppression that fits in a 4 mm × 1.6 mm footprint.
LTCC filters come in a SMT form factor and are similar This size is significantly smaller than half a wavelength,
to multilayer capacitors, in which multiple layers of very enabling integration in both plank and planar architec-
thin ceramic tape are printed with different passive ele- tures.
ments and then stacked together to prevent substrate
warping. Prototypes of LTCC technology for mmWave CONCLUSION
applications are being developed, potentially offering The timeline for delivering mainstream 5G communi-
a method of including both filters and antennas on the cations is getting tighter, and determining the appropri-
same component with a very small footprint. Unfortu- ate RF filtering for 20 GHz and above is a fundamental
nately, since the metal coatings are screen printed, di- issue. 5G systems require filters with high percentage
mensional precision is not as high as other thin film so- bandwidths, good selectivity and excellent tempera-
lutions and the unpolished substrate can lead to high ture stability in compact packages. In order to acceler-
losses.6 Plus, suppression is generally limited to 30 dB ate time-to-market, developers are seeking established
and below. solutions, such as waveguide and microstrip filters, that
Another type of SMT filter assembly is single-layer have long been used in the satellite, radar and broad-
microstrip, which is printed with distributed transmis- casting industries.n

BPF, 26.5 to 29.5 GHz, S21 & S11

References
0 1. T. S. Rappaport, “Millimeter Wave Cel-
lular A Road to 5G,” IEEE International Con-
–10 ference on Communications,” June 2013,
https://wireless.engineering.nyu.edu/pre-
–20 sentations/icc2013.pdf.
2. K. Sakaguchi, T. Haustein, S. Barbarossa,
Magnitude (dB)

–30 E. C. Strinati, A. Clemente, G. Destino, A.

Pärssinen, I. Kim, H. Chung, J. Kim, W. Keus-
–40 gen, R. J. Weiler, K. Takinami, E. Ceci, A. Sa-
dri, L. Xain, A. Maltsev, G. K. Tran, H. Oga-
–50 wa, K. Mahler and R. W. Heath Jr., “Where,
When, and How mmWave is Used in 5G and
–60 Beyond,” IEICE Transactions on Electronics,
Vol. E100-C, No. 10, October 2017.
–70 3. “5G NR Primer for Amplifier and Filter
Design,” National Instruments, September
–80
10,000 14,000 18,000 22,000 26,000 30,000 34,000 38,000 2018, www.awrcorp.com/news/article/5g-
Frequency (MHz)
new-radio-primer-amplifier-and-filter-de-
sign-now-available.
s Fig. 5 Repeatability of Knowles Precision Devices 28 GHz SMT single-layer 4. J. Conrood, “Substrate Integrated
microstrip filters, using 100 samples without tuning. Waveguides (SIW) Limits and Capabilities,”
Rogers Corp., January 2018, www.youtube.
com/watch?v=I9A66sUqaTk.
TABLE 2 5. T. Pensala, “MEMS & Thin Films Manu-
facturing Platform for Mm-wave and 5G
COMPARING SAW/BAW FILTER PERFORMANCE WITH SMT MICROSTRIP RF Components,” Eemeli Workshop, May
Typical SAW/ SMT Microstrip 2015, www.vttresearch.com/Documents/
Performance Goals BAW Performance Performance Eemeli/20th%20Eemeli/05%20Pensala%20
(< 6 GHz) (> 20 GHz) Eemeli%2021.5.2015%20mm-wave%20
MEMS%20thin%20film%20manufactur-
Low Insertion Loss < 3 dB < 3 dB
ing%20platform%20VTT%20TPensala.pdf.
> 30 dB > 50 dB 6. N. Friedrich, “The Do’s And Don’ts Of
Excellent Rejection
out of band out of band LTCC,” Microwaves & RF, September 2013,
www.mwrf.com/passive-components/do-s-
Broad Bandwidth < 100 MHz 3 GHz
and-don-ts-ltcc.
Small Size ~9 mm2 < 9 mm2
Good Temperature Stability ~3 ppm/ºC ~3 ppm/ºC

7
Designing a Narrowband
28 GHz Bandpass Filter for
5G Applications
David Vye and John Dunn
NI AWR Group
Dan Swanson
DGS Associates
Jim Assurian and Ray Hashemi
Reactel Inc.
Philip Jobson
Design Consultant

This article examines the factors driving the physical, electrical and cost constraints for 5G filters.
To address these challenges, a narrowband filter design methodology using classic filter network
theory, parameterized electromagnetic (EM) simulation and port-tuning techniques is presented.
The approach is demonstrated using the NI AWR Design Environment platform to develop a
narrowband 28 GHz bandpass cavity filter targeting mmWave backhaul applications.

5
G will increase network capacity, reduce la- the most cost-effective and versatile solution to connect
tency, and lower energy consumption through 5G base stations to the core network. Filters developed
a number of innovative technologies aimed at for the wireless backhaul application will face cost and
enhancing spatial and spectral efficiency. The volume challenges that must be considered early in the
use of carrier aggregation, mmWave spectrum, base design stage.
station densification, massive MIMO and beamforming
antenna arrays will combine to support the goals of 5G DESIGN APPROACH AND FILTER
communications at the cost of more signals operating SPECIFICATIONS
in close spectral and spatial proximity. These enabling Ideal filter responses are well defined by math func-
technologies place new demands on the filters required tions. This has led to the development of numerous
to mitigate signal interference across a dense network of commercial synthesis tools that can generate circuits for
base stations and mobile devices. an exact filter response based on ideal element values;
however, the parasitic behavior of the filter components
5G APPLICATIONS must be considered early in the design stage. For this
5G will be deployed in stages to address three main reason, synthesis is excellent for accelerating the initial
thrusts; enhanced mobile broadband (EMBB), massive design phase and generating mathematical filter solu-
machine-type communication (mMTC) and ultra-reliable tions to serve as a starting point to define ideal lumped
and low-latency communication (URLLC) for remote sens- or distributed networks. Synthesis, however, is also lim-
ing and control for medical and autonomous vehicle ap- ited in its ability to generate a physically-realizable filter.
plications. On the infrastructure side, densely-populated In this case, the synthesis tool provides critical coupling
urban environments will utilize the mmWave frequency coefficients and external Q targets, but the ideal electri-
spectrum for higher data rates. Wireless backhaul is likely cal design has limited usefulness.
www.mwjournal.com/articles/32052
8
 Synthesis tools such as iFilter™ filter synthesis within filter’s optimal response. For a lossless Chebyshev filter,
NI AWR software can perform the math to produce ideal the optimal behavior is an equal ripple insertion and re-
LC filters and precise distributed designs such as edge turn loss response in the pass band. Thus, if the opti-
coupled, hairpin, interdigital and combline, based on mizer can consistently find this equal ripple response,
ideal distributed microstrip and stripline models. Figure optimization can reliably be used. The optimizer that is
1 shows several types of narrowband filters that can be used in this project is available as an add-on module
synthesized based on microstrip technology with ideal to the NI AWR Design Environment platform using the
distributed models that do not incorporate manufactur- software’s API COM interface to integrate fully with Mi-
ing limits and tolerances. Addressing these uncertain- crowave Office circuit design software and AXIEM and
ties can be very difficult without a process for converting Analyst™ EM simulators.
ideal designs into physically-realizable ones.
The method used in this design is based on a tech- Designing a Physically-Realizable 5G Filter
nique introduced by Dishal and adopted for use with The design methodology follows a set of well-defined
modern circuit and EM simulation by co-author Dan steps that scale for the desired frequency and bandwidth
Swanson. EM modeling is used to efficiently determine (see Figure 2). The process starts with specification of
three fundamental filter properties: the unloaded Q of the filter requirements, including bandwidth, passband
the internal resonators, the coupling between two ad- return loss and stopband rejection, from which the fil-
jacent resonators and the external Q of the two resona- ter order is determined and the lowpass Chebyshev
tors that form the input and output connections. Para- parameters are determined and scaled to the required
metric studies with EM analysis are critical in modifying frequency.
the physical structure in order to obtain specific values An EM model of a single resonator is built, and its
for these filter properties, which are determined by the length for a desired resonant frequency and unloaded
Dishal method. Port tuning is then applied using circuit Q is determined. Additional EM models are created to
simulation and optimization with ideal lumped-element generate the coupling coefficient and external Q curves
components, specifically, capacitors that are placed in that guide the determination of key physical dimensions
strategic locations. Port tuning is used to guide adjust- such as resonator spacing and tap height. These individ-
ments that must be made to the final physical design. ual components are then assembled, and port tuning is
used to tweak the design for the optimum equal ripple
Design by Optimization response using optimization.
General purpose optimizers are not particularly effi- Narrowband Bandpass Filter Design
cient for filter design unless they are able to take ad- The filter is designed with a center frequency of 28
vantage of the mathematical foundation that defines a GHz, (3GPP band n257). The construction is based on a

Interdigital Tapped Combline

d

Hairpin Optimum Distributed Bandpass

Short Stub Bandpass b

Edge Coupled L

25.5 28.0 30.5

s Fig. 1 Several types of narrowband filters that can be synthesized based on D

microstrip technology.
Optimal Z0 is Approximately 77 Ω
Optimal d/D is Approximately 0.33
Resonator Coupled Pair Tapped Port Electrical Length is Typically 30º to 60º
Spec
Design Design Resonator Tune/Optimize

s Fig. 3 Coaxial resonator cross

s Fig. 2 Steps for designing a physically-realizable 5G filter. section.

TABLE 1
IDEAL CHEBYSHEV LOWPASS FILTER RESPONSE BASED ON CUTOFF FREQUENCY NORMALIZED TO 1 Hz
N g0 g1 g2 g3 g4 g5 g6 g7 g8 g9 g10 ∑ g1- gN
2 1.0000 0.6682 0.5462 1.2222 1.2144
3 1.0000 0.8534 1.1039 0.8534 1.0000 2.8144
4 1.0000 0.9332 1.2923 1.5795 0.7636 1.2222 4.5727
5 1.0000 0.9732 1.3723 1.8032 1.3723 0.9732 1.0000 6.4989

9
 TABLE 2
FREQUENCY RESPONSE FOR SEVERAL DIFFERENT
RESONATOR LENGTHS
–30
R length F0 (GHz) Qu

–40 0.07883 in. 27.9535 1975.2

m1:
–50 0.07878 in. 28.022 2040.6
27.9536 GHz
–31.16 dB 0.07773 in. 28.043 1978.1
–60

–70 In addition to the cavity dimensions, another early

27.8 27.9 28.0 –30
28.1 28.2 S21 DB(|S2,1|)
Rlength=0.07878
concern is determining the cross-sectional dimensions
Frequency (GHz)
DB(|S2,1|)
Rlength=0.07883 of the resonator post. The resonator post cross section
DB(|S2,1|)
–40 Rlength=0.07873 in relation to the outer cavity wall determines the char-
s Fig. 4 Coaxial resonator EM model.
–50
acteristic impedance of the resonator. For a coaxial reso-
nator, literature indicates an optimum resonator charac-
S21 –60
DB(|S2,1|) teristic impedance of around 77 ohms, as determined
–30 Rlength=0.07878
DB(|S2,1|)
–70
Rlength=0.07883
by the resonator cross-section, resulting in a post-to-
–40
DB(|S2,1|)
27.8
Rlength=0.07873 27.9 28.0 28.1 28.2 cavity width ratio of about 33 percent (see Figure 3). In
Frequency (GHz) this case, the optimum unloaded Q (Qu) is not achieved
–50
m1: 28.0214 GHz, –30.85 dB due to physical constraints. A coaxial transmission line
–60
m2: 28.0483 GHz, –31.17 dB calculation approximates the resonator Zo ~46 ohms.
m3: 27.9539 GHz, –31.17 dB
–70 Resonant Frequency and Qu Simulation
27.8 27.9 28.0 28.1 28.2 With the cavity and resonator cross-section dimensions
Frequency (GHz)
determined, an EM model of a single resonator is defined
s Fig. 5 m1:
EM28.0214
resonator simulation
GHz, –30.85 dB results for different with the resonator length parameterized so that the reso-
resonatorm2: 28.0483 GHz, –31.17 dB
lengths. nant frequency can be controlled, as shown in Figure 4.
m3: 27.9539 GHz, –31.17 dB
The EM model uses two coaxial feed structures that are
single in-line cavity using an interdigital arrangement to
loosely coupled to the waveguide cavity to act as the input
achieve 30 dB of rejection at 800 MHz off center fre-
and output ports. The frequency response for several dif-
quency and an in-band return loss of 20 dB. This sets the
ferent resonator lengths (see Figure 5), demonstrates that
in-band ripple at 0.044 dB. From these specifications,
the resonant frequency increases with a shortening of the
the expected insertion loss and the necessary filter or-
resonator.
der are determined.
The Qu of an individual resonator is calculated from
The mathematical foundation for an ideal filter re-
the simulated time delay and insertion loss using Equa-
sponse is well established, with parameter values de-
tion 2.
rived for an ideal lowpass Chebyshev filter response
based on a cutoff frequency normalized to 1 Hz (see
Table 1). Once the prototype ripple level is determined 10IL(dB)/20
Qu = π f0 t d (2)
from the desired in-band return loss, the filter order N 10IL(dB)/20 − 1
can be estimated based on the desired stopband rejec-
For lowpass Chebyshev parameters, center frequen-
tion, as shown in Equation 1. A fifth-order filter is need-
cy, and simulated Qu are shown in Table 2. An expected
ed to achieve the desired selectivity and bandwidth.
insertion loss of approximately 0.25 dB is calculated for
Rejection(dB)+RtnLoss (dB)+6 the entire filter at mid-band using Equation 3.
N>
( )
(1)
20log10 S + S2 − 1 4.343 f0 N
Loss ( f0 ) = ∑ gi (dB) (3)
Δf Qu i=1
Where:
Rejection = Stopband Insertion Loss
Where:
RtnLoss = Passband Return Loss
S = Rejection Bandwidth/Filter Bandwidth ∆f is the equal ripple bandwidth of the filter
Design Details Qu is the expected average unloaded Q for the resona-
The design is based on an interdigital configuration tors
made up of coupled resonators with the open ends on gi are the normalized lowpass filter element values, cal-
a substrate or cavity alternately pointing in opposite di- culated for a given ripple in the Table 1.
rections. The length of the resonators determines the An accurate accounting of manufacturing factors
resonant frequency and the coupling between resona- such as surface roughness and plating details are miss-
tors is controlled by their separation. The width of the ing from the model used for EM simulation, so the in-
housing, for a cavity filter should be λ/4 at the operating band insertion loss will likely be higher than this initial
frequency. estimate. The model approximates 80 percent of the
10
ideal conductivity as a starting point. The quality of the The coupling between the resonators results in a dis-
silver plating is very process dependent. The measured placement ∆f of the resonant frequencies, which is
data from the manufactured and tested filter can be known as the coupling bandwidth. By dividing the cou-
used to adjust model conductivity information. pling bandwidth by the ripple bandwidth of the filter,
the normalized coupling coefficient is obtained. The
Coupling Coefficient Simulation
From the Chebyshev lowpass filter parameter values
∆f: 1.215 GHz, Spacing = 0.85 in.
(g in Table 1) the external Q and the coupling coeffi- 0
cients (kij) for the resonant pairs are calculated based on m1 m2
Equations 3 and 4, respectively, using a 2.85 percent –10
bandwidth.
–20
DB(| S(2,1)| )
f g g g g Coupling
Qex = 0 0 1= 0 1 –30
f2 − f1 BW
–40

Kij =
( f2 − f1 ) = BW
(4) –50
f0 gig j gig j f0 Shifted to 28.074 GHz
–60
f1 + f2 f2 − f1 27.0 27.5 28.0 28.5 29.0
f0 = BW =
2 f0 Frequency (GHz)
(a)
f1 = bandpass filter lower equal ripple frequency
∆f: 0.38 GHz, Spacing = 0.125 in.
0
f2 = bandpass filter upper equal ripple frequency m3 m4
–10
f0 = bandpass filter center frequency
–20
DB(| S(2,1)| )
BW = percentage bandpass Coupling
–30
gi = Prototype element value for element i
–40
Note: Equations assume Qu is infinite.
–50

These calculated values provide the targets for the –60

physical design. The next step is to build the coupling 27.0 27.5 28.0 28.5 29.0
coefficient design curves using an EM model of the Frequency (GHz)
coupled resonators in order to determine the necessary
spacing. m1: 27.466 GHz Ref m3: 27.817 GHz Ref
m2: 1.2145 GHz Delta m4: 0.814 GHz Delta
Two resonators based on the initial resonator study (b)
are enclosed in a metal cavity and loosely coupled to
the input and output ports, as shown in Figure 6. The s Fig. 7 Simulated transmission characteristics of two
resonators enclosed in a metal cavity and coupled to input and
resonators are identical and resonate at frequency f0. output ports: cavity spacing equals 0.085 in. (a) and 0.125 in. (b).

8
SUBCKT
ID=S1
NET=”ANA_COUPLING” 6
XFMR GSP2=0.125 XFMR |Y(1,1)| Coupling
ID=X1 ID=X2 |Y(2,2)| Coupling
Port N=N1 N=N1 4
P=1 1 1:n1 3 1 2 3 n1:1 1 Port
Z=50 Ohm P=2
Z=50 Ohm 2
2 4 4 2
0
CAP CAP 27.0 27.5 28.0 28.5 29.0
ID=C1 ID=C2 Frequency (GHz)
C=C1 pF C=C2 pF
N1=35
C1=0.000614693773294939
C2=0.00057077657585845
1

28.0
Frequency (GHz)
2

s Fig. 8 The f0 shift is addressed through

s Fig. 6 Ports introduced for port tuning of coupled resonators. tuning c1 and c2 values using optimization.

11
normalized coupling coefficient divided by the center cy. It can also be seen that the center frequency be-
frequency provides the Chebyshev lowpass coupling tween the peaks in Figure 7a are shifted upward 74 MHz
coefficient. The resonant frequency occurs at the mid- for the case where the resonator spacing is 85 mils. The
point between the two peaks, as shown in Figure 7. admittance vs. frequency for these two ports is simu-
The more closely the resonators are spaced, the far- lated and the capacitor values are optimized to zero out
ther apart are the resonant peaks; this corresponds to the admittance at 28 GHz, which re-centers the resonant
higher coupling. As the resonators move farther apart frequency of the coupled pair (see Figure 8). The impact
the coupling gets progressively weaker and the two of the small amount of capacitance added or removed
peaks merge together at the original resonant frequen- in order to center the coupled resonance can then be re-

Coupling Data
130
125
y = –708125x3 + 85686x2 – 4329x + 169.36 –26
120 –27
Resonator Spacing (mils)

115 –28
–29
110

S21 (dB)
–30
105 3 dB
–31
∆f3dB
100 –32
95 –33
Coupling Data –34
90
Poly. (Coupling Data) –35
85 –36
f0
80
Frequency (GHz)
0.010 0.020 0.030 0.040 0.050
Coupling Coefficient

s Fig. 9 Inverse relationship between the amount of coupling s Fig. 10 External coupling is found by measuring the 3 dB
between resonators and their spacing. bandwidth of the resonance curve.

2 Port Network with Tuning Cap SUBCKT

ID=S1
Implemented with Circuit Schematic NET=”ANA_Qex”
TapHeight=0.031
Port
P=1 1 2
Z=50 Ohm
1
2 CAP
ID=C1
C=0.0015 pF

Tap Location is Swept

in z-Direction

s Fig. 11 Single resonator EM model includes a coaxial feed with a parameterized tap feed height to adjust external Q.

Qex = 2πxf(GHz) td (ns) = 2πx27.734x0.7215 = 31.43

4 4
0.8 40
m1
m1
0.6 30
GD(1,1) (ns) Qex m1: Maximum 27.734 GHz |Eqn(Qez)| m1: Maximum 27.737 GHz
0.7215 ns Output Equations 2 31.43
0.4 20

0.2 10

0 0
26 27 28 29 30 26 27 28 29 30
Frequency (GHz) Frequency (GHz)

s Fig. 12 Simulated reflected time delay for a given tap height.

12
placed by adjusting the resonator length to add/remove By parameterizing the spacing and tweaking the
an equivalent amount of capacitance. resonator length through port tuning, a curve relating
coupling coefficients to very accurate resonator spac-
Calculating Kij Curves From Parametric EM Analysis ings based on EM analysis can be calculated. From this
Dividing the normalized coupling coefficients by the curve the spacing necessary to achieve a required cou-
28 GHz center frequency provides the coupling coeffi- pling is determined. The curve in Figure 9 shows the
cients that are needed to match up to the lowpass Che- anticipated inverse relationship between the amount of
byshev parameter values. coupling between resonators and their spacing.
From Kij calculations:
Parametric Modeling of the Tapped Resonator
[K1,2],[K4,5] = 0.02466 The next step is to determine the physical details of
[K2,3],[K3,4] = 0.01812 the tapped resonators that provide the input/output to
Coupling bandwidth [1,2][4,5] = the filter. The external coupling is found by measuring
690 MHz the 3 dB bandwidth of the resonance curve denoted by
∆f3dB (see Figure 10). The external Q is Qext = Qload-
Coupling bandwidth [2,3][3,4] = ed = f0 / ∆f3dB. It is also possible to determine the ex-
507 MHz ternal Q by measuring the group delay of S11.
(= Kij × 28GHz) A parameterized EM model including a coax feed
that taps into a single resonator is created and the dis-
Qex Data tance from the bottom of the housing to the center of
32
y = –0.0007x3 + 0.0831x2 – 3.94x + 88.858
the coax tap is parameterized so that it can be adjusted
31 to different heights to achieve the external Q calculated
30 from the Chebyshev lowpass parameter. A lumped port
29 is also placed between the resonator and tuning screw to
Tap Height

28 support port tuning for addressing shifts in the resonator

27 frequency due to the tap (see Figure 11).
26 EM analysis of the tapped resonator provides the time
25 delay response as a function of frequency for different tap
Qex Data
24 heights (see Figure 12). The time delay response is used
Poly. (Qex Data)
23 to derive the external coupling. Parametric simulation en-
22 ables the generation of an external Q vs. tap height curve,
20 25 30 35 40 from which the tap height necessary for the required Qex
Qex can be directly chosen (see Figure 13).
s Fig. 13 Qex vs. tap height based on a parameterized swept Port Tuning
EM analysis of reflected time delay.
Parameterization is used to sweep values and gener-
ate the individual components that are combined to re-
produce the entire filter and finalize the design through
port tuning using the equal ripple optimization routine.
While today’s EM simulators are quite fast and powerful,
EM simulation times for low-order filters is still on the or-
der of minutes or tens of minutes. Port tuning moves the
optimization process from the EM domain to the circuit
s Fig. 14 Adding a port at each resonator enables tuning theory domain, where simulation times are much faster.
the frequency of each resonator and the coupling between Adding a port at each resonator enables rapid tuning of
resonators.

Fn SWPFRQ
Fo Port Tuned Analyst Model
IND ID=FSWP1
ID=L1 Values=(27.52, 27.668864, 27.920497, 28.169457, 28.328) 0
L=0 nH SUBCKT –10
Port 1 ID=S1 7 Port
P=1 NET=”N5 INTDIG” P=2 –20
Z=50 Ohm 2 6 Z=50 Ohm
3 4 5 –30
CR1=0
CR2=0 CAP CAP CAP CAP –40
CR3=0 ID=C6 ID=C7 ID=C8 ID=C9 –50
CR4=0 C=CC1 1F C=CC2 1F C=CC3 1F C=CC4 1F
CR5=0 –60
26.5 27.0 27.5 28.0 28.5 29.0 29.5
CC1=0 Frequency (GHz)
CC2=0
CAP CAP CAP CAP CAP DB(|S(1,1)|) Schematic 1
CC3=0 ID=C1 ID=C2 ID=C3 ID=C4 ID=C5
CC4=0 C=CR1 1F C=CR2 1F C=CR3 1F C=CR4 1F C=CR5 1F DB(|S(2,1)|) Schematic 1
(a) (b)

s Fig. 15 Coupling parameter simulation (a) and resulting S-parameters (b).

13
each resonator and the coupling be- Screw Length
tween resonators (see Figure 14). = 0.015 in.
0.078 0.082
With each resonator loaded by a 50 0.082
0.078 0.12
0.078 0.02
ohm port, the raw coupling between 0.02
resonators (not coupling coefficients
per se) is simulated and the S-param-
eter variation across the simulation 0.05
0.1
domain is extremely smooth (see Fig-
0.226
ure 15). In fact, for a narrowband filter, 0.388
0.07
only five to 10 discrete frequencies 0.550
across the simulation domain are re- 0.70

quired for the circuit simulator to gen- 0.77

erate a smooth frequency response
plot through interpolation.
With port tuning, the resulting ca- s Fig. 16 Final dimension of the filter design, derived from port tuning.
pacitance values reveal the tuning
requirements for the 3D EM model. Both positive and
negative capacitance values can be used in circuit simu-
lation. For the resonator tuning (port to ground capaci-
tors), a negative capacitance value indicates that the
resonator (EM model) is tuned too low. Positive capaci-
tance represents a resonator that is tuned too high. For
adjusting the coupling (port to port capacitors), a posi-
tive series capacitance indicates that the coupling is too
strong in the EM model (the resonators are too close).
The process is repeated until the capacitances be-
come sufficiently small. Convergence is guaranteed if
the changes are not too large. Once resonator sensitivi-
ties (kHz per mm) are known, capacitance values can be (a)
converted into physical changes of the structure. Figure
16 shows the dimensions for the final design, derived 28 GHz Narrowband Filter Response
0
from the port tuning.
–10
SIMULATED VS. MEASURED RESULTS
From this design, the filter manufacturer (Reactel Inc.) –20
built and tested the cavity filter, shown without the cover
in Figure 17a. The frequency response of the measured –30
filter and simulated model is shown in Figure 17b. As
–40
designed, the target response is achieved with moder- DB(|S(2.1)|) Measure Data
ate screw tuning. More precise tuning would better rep- –50
DB(|S(1.1)|) Measure Data
licate the optimized, simulated result. DB(|S(1.1)|) Simulation
DB(|S(2.1)|) Simulation
Manufacturing Tolerances and Yield Analysis –60
26.5 27.5 28.5 29.5
Modern CNC machines offer 0.0002 in. tolerances, not Frequency (GHz)
(b)
including tooling and fixturing. The relationship between
3D EM model and port tuning capacitors (resonators and s Fig. 17 Pre-plated cavity (a) and simulated vs. measured
coupling) can be used to perform yield analysis using the results (b).
circuit simulator, allowing physical tolerances from manu- filter models. EM tools continue to mature and add
facturing process to be translated into capacitor toleranc- capabilities/speed, making it practical to include them
es for yield analysis. Yield analysis of microwave circuits is in an optimization loop. This technique has been used
often done with a Monte Carlo-type analysis with a large to address the challenge of designing highly sensitive
number of iterations. Running these iterations in the EM mmWave filter designs.n
domain is prohibitive, but if the computed sensitivities
convert a capacitance to a physical dimension, yield opti-
mization is possible through the circuit simulation.

CONCLUSION
A practical design method that is independent of
filter type/construction has been demonstrated, show-
ing a robust equal ripple filter optimization that is a fast
and intuitive alternative to design by synthesis and an
efficient approach for port tuning complex EM-based

14
 WHITEPAPER

Tolerance and Size Analysis for mmWave Filter Manufacturing

One of the questions we get asked regularly is • Center Frequency: 26GHz

‘why not just integrate a filter in the board stack?’. • 3dB Bandwidth: 4x800MHz
• Insertion Loss: 2dB
Besides overall filter performance concerns, our • Stopband 1: -35 dB within DC-22 GHz
answer to this comes in two parts: First there • Stopband 2: -35dB within 30-68 GHz
are manufacturing tolerances to consider, and
second there is size. Impact of manufacturing tolerance on
filter performance
Manufacturing tolerance for any filter will
contribute to a filters performance and can make For manufacturing tolerances we used our own
for some interesting discussions about what was experience in fabricating filters across a range
planned for vs the realized filter performance. of substrate materials. For the comparison we
At mmWave tolerance impacts can become selected three filter implantation approaches –
significant and can potentially impact the total An integrated Microstrip approach, an integrated
cost of an implementation. Stripline approach and one of our own Thin Film
Dielectric component designs.
Size is also a cost of implementation
consideration – allocating board space or layers Notes on tolerance analysis
to a filter implementation can take up areas that The approach we take is a simplified worst case
can be used for other functionality or devices and analysis. Discussion in the public domain [1]
may further complicate implementation. use a monte-carlo based approach to reach
essentially the same conclusion for Microstrip
To look at these factors we examined different on PCB. In general we would observe that the
26GHz filter manufacturing approaches with the tolerances used in these prior analyses is very
following design goals: aggressive compared to the typical tolerances
that can be held in production.

tolerance_mmWave 06/25/19
knowles.com 1
15
 WHITEPAPER

Table 1. Manufacturing tolerances used in this work (worse case analysis) Table 2. Manufacturing tolerances used in [1] (monte carlo analysis)

Tolerance Stripline Microstrip Std-Dev Microstrip on 3.4 low loss laminate

on RO3003 on RO4350B Thickness 15um, 0.6mil
Thickness 30mil ±1.5mil 10mil ± 1mil Line width 8um, 0.3,mil
Etching Tolerance ±0.5mil ±2mil Permittivity 3.4 ±0.02
DK Tolerance 3 ±0.04 3.66 ±0.05

Microstrip any further since adding complexity would drive

As first pass we had a look at a Microstrip up insertion loss and sensitivity to tolerances.
filter. We found a close match to the design
Figure 2. Simulated S21 from Microstrip Implementation of
targets above was to implement a 6th order
26GHz BPF, with limit lines
edge coupled topology on 10mil of RO4350B.
Metallization was 2oz Electrodeposited (ED) S21, RO4350B Microstrip Filter
0

Copper. -10

Figure 1. Microstrip Implementation of 26GHz BPF -20

-30
Magnitude (dB)

-40

-50

-60

-70

-80
20 22 24 26 28 30 32 34
Frequency (GHz)

To look at tolerance impact on this hypothetical

Microstrip filter we ran a worst case analysis
over 8 variations of the design, using variations
Due to the inherent transmission zeros produced of substrate thickness, etching tolerance and
by the edge coupled Microstrip filter the nominal dielectric constant tolerance.
design passes the close-in rejection requirements
with margin, and while the passband shows less For this design we selecting the following
margin the design could be adjusted to provide manufacturing parameters, knowing they would
some additional space. The high side harmonic be considered reasonably tight without being
response is less than favorable, but for the sake outside the realms of available manufacturing
of this analysis we decided not to tune the design capability:

tolerance_mmWave 06/25/19 2
16
 WHITEPAPER

Figure 4. Stripline Implementation of 26GHz BPF

• RO4350B Thickness = 10mil ± 1mil
• Etching Tolerance = ±2mil
• DK Tolerance = 3.66 ±0.05

Using a simple worst case analysis we

observe a significant shift in the filters low side
performance for both rejection and 3dB point.
The 35dB low side rejection point is shifting by
over 2GHz. Based on this simplified analysis,
unless significant margin is added to the target approach. For this simulation the following design
Figure 3. Simulated S21 over 8 manufacturing runs of comes close to the target specifications we are
Microstrip Implementation of 26GHz BPF, with limit lines working with:
Microstrip RO4350B Tolerance
0

-10
• 7th Order Hairpin
-20 • 30mil RO3003 Top and Bottom
-30
• 0.5oz Rolled (RA) Copper
Magnitude (dB)

-40

This hairpin stripline design does not have quite

-50

-60

-70
as steep a skirt as the edge coupled microstrip

Figure 5. Simulated S21 from Stripline Implementation of

-80
20 22 24 26 28 30 32 34
Frequency (GHz)

26GHz BPF, with limit lines

specification such a filter would be unusable 0

S21, RO3003 Stripline Filter

in manufacturing – being part of the PCB this -10

variation over tolerance would render the entire -20

board out of specification before performance -30

Magnitude (dB)

variations from other components or due to -40

factors like temperature stability come into play.

-50

-60

Stripline
-70

-80

Next up we will take a look at a stripline

20 22 24 26 28 30 32 34
Frequency (GHz)

tolerance_mmWave 06/25/19 3
17
 WHITEPAPER

implementation and we find the same harmonic Figure 7. Tolerance simulations for the Stripline Filter
performance challenges on the high side of the
stop-band as seen with the microstrip filter.
Stripline RO3003 Tolerance
0

-10

Figure 6. Comparison of nominal design performance, -20

Microstrip and Stripline -30

Magnitude (dB)
-40

Nominal Filter Comparison
0 -50

‐10 -60

‐20 -70

‐30 -80
20 22 24 26 28 30 32 34
Magnitude (dB)

Frequency (GHz)

‐40
RO3003 Stripline Filter
RO4350B Microstrip Filter

Knowles Precision Devices Catalog Filters.

‐50

‐60

‐70
For comparison we will look at the simulated
‐80
tolerance shift in our one of catalog filters – the
20 22 24 26 28 30 32 34 36 38 40

B259MC1S. We can’t get into topology and

Frequency (GHz)

material details here, but we can show nominal

For tolerance analysis we took a similar
simulated design performance and simulated
approach, looking at 8 variations across substrate
tolerance response to the same three variables.
thickness, etching tolerance and dielectric
material tolerance. Figure 8. Nominal design performance – B259MC1S.
S21, Knowles B259MC1S
0

• RO3003 Thickness = 30mil ±1.5mil -10

• Etching Tolerance = ±0.5mil -20

• DK Tolerance = 3 ±0.04 -30

Magnitude (dB)

-40

From the worst case analysis we observe a

-50

smaller shift in filter performance. However,

-60

-70

margin would still need to be added to the design -80

to allow for the variation in frequency response.

20 22 24 26 28 30 32 34
Frequency (GHz)

For example on the low side the 35dB point is

varying by ~1GHz.
If we subject the simulated version of the catalog

tolerance_mmWave 06/25/19 4
18
 WHITEPAPER

filter to the same worst case analysis across It is also a potential space saver, which is an
just three design variables we can see a much aspect we can look at next.
smaller sensitivity to manufacturing tolerances.
Filter Size Comparison.
Figure 9. Knowles B259MC1S – simulated performance
Let’s take a look at the dimensions of each
variation across 3 variables
design as simulated:
Knowles B259MC1S Tolerance

• Microstrip Design Size: 505 x 170 x 10mils, or

0

-10

-20 12.8 x 4.3 x 2.5 mm

-30
• Stripline Design Size: 560 x 160 x 60mils, or
Magnitude (dB)

14.2 x 4.0 x 1.5mm

-40

-50

-60
• Knowles B259MC1S Size: 216 x 90 x 64mils,
-70 or 5.5 x 2.3 x 1.6 mm
-80
20 22 24 26 28 30 32 34
Frequency (GHz)

Looked at another way, we can look at the

Simulated 35dB attenuation shift at the low end is nominal performance comparison again and
approximately 130MHz and the simulated filters overlay the filter surface area to scale.
do not threaten the stop bands. We observe a Figure 10. Nominal design performance and Filer Area
small shift vs the static passband. Also given
that these devices are discrete surface mount 0
Nominal Filter Comparison

Knowles B259MC1S, 5.5x2.3mm

components (rather than implemented within the -10

Stripline, 14.2x4.0mm
PCB) and we screen the filters by 100% RF test, -20

Microstrip, 12.8x4.3mm
any devices that did not meet the performance -30
Magnitude (dB)

goals would not ship to the end user.

-40 Knowles Thin Film Ceramic Filter
RO3003 Stripline Filter
RO4350B Microstrip Filter

-50

-60

Adding fully tested surface mount filters with

-70

known repeatable performance is a cost saver -80

20 22 24 26 28 30 32 34

over integrated board level implementations in Frequency (GHz)

microstrip or stripline when one considers the

total cost in rejecting non-conforming board level Considering performance for footprint, a compact
assemblies. SMT filter takes up less than a quarter of the
space that would be allocated to the Microstrip

tolerance_mmWave 06/25/19 5
19
 WHITEPAPER

and Stripline filters built to the same target • If you are implementing filters on (or in)
specification. board, consider the lot to lot repeatability
and allow for this in your target
Looking next at where the filters actually fit into specifications.
a system we need to remember that antenna
spacing in phased array systems will be of • Guard bands can eat into available
the order of λ/2, where λ is the free space bandwidth, undoing some of the gains for
wavelength of the radiation that the antenna will using mmWave frequencies originally.
transmit and or receive.
• Tolerance is a cost to implement driver
At 26GHz λ/2 is approximately 5.7mm, smaller if one considers the cost of rejection no-
than the largest dimension of the two PCB conforming boards, and the physical size of
implementations we looked at. the filter implementation is another potential
cost consideration
Figure 10. Longest dimension compared to λ/2 at 26GHz

Largest Dimension, Compared with λ/2 spacing at 26GHz
16.00

Stripline
14.00

Microstrip

12.00 References
10.00
Dimension (mm)

8.00
[1] On mm-wave filters and requirement impact,
6.00 26GHz λ/2 (mm)
KPD 3GPP TSG-RAN WG4 meeting #85, R4-1712718
4.00

2.00

0.00

Observations

• At mmWave frequencies manufacturing

tolerances can make significant
contributions to a filters performance.

tolerance_mmWave 06/25/19 6
20
Global 5G Rush
But No Global 5G Handsets
Ben Thomas
Qorvo, Greensboro, N.C.

W
e are currently witnessing a global rush to ranges: sub-6 GHz (FR1) and mmWave above 24
5G. Nations, mobile operators and hand- GHz (FR2). South Korea, Britain, Italy and Spain,
set manufacturers are all vying to be the among others, raised billions of dollars in spec-
first to deliver the next-generation of cellular con- trum auctions during 2018, and the U.S., Chi-
nectivity—or at least to get in the game early. na, Japan and Australia will hold auctions and
Worldwide, there are robust plans for rapid 5G make allocations in 2019. Operators in many
deployment, especially in regions where the countries, including the U.S., plan to start roll-
wide bandwidth provided by new 5G bands ing out 5G services in 2019, and several major
can provide significantly higher data rates for handset makers have said they will produce
consumers. Indeed, it is this access to New 5G phones supporting these services.
Radio (NR) bands with the refarming of Overall, these initiatives are driving
existing LTE bands that provide the toward widespread 5G cover-
greatest impact on data rates (see age in developed countries
Figure 1). Unlike the 3G to LTE by 2021.
transition, the change in un- Yet the global drive
derlying 5G specifications to 5G does not mean
provides only a modest that we will see
data rate improvement. global 5G hand-
This explains why, to fa- sets. In contrast
cilitate fruitful 5G deploy- to the situation
ment, countries are rapidly allocating new with LTE, it may
spectrum in both of the newly designated not be feasible
or cost-effective to
build global 5G hand-
 19% sets that support roam-
ing across 5G networks
worldwide. Instead, 5G will
likely drive the handset market
in the opposite direction—toward
greater regionalization.

5G BANDS NOT GLOBAL

3G LTE 5G
First, new “global” FR1 bands
s Fig. 1 5G requires added spectrum, as the (n77, n78 and n79) are, in practice,
specifications provide only modest data rate anything but global; in many cases,
improvements. countries are allocating different
www.mwjournal.com/articles/31757
21
 Further, leading manufacturers have
Standards chosen to participate in the Global
n78 3.3 to 3.8 GHz
Bands 3GPP
5G n77 3.3 to 4.2 GHz Certification Forum’s (GCF) interop-
erability certification, which provides
Korea
China
benefits for roaming with LTE handsets.
Japan The common GCF practice is to cer-
Country U.K. tify operation over the full bandwidth
Allocation EU of the designated band, which begs
U.S. TBD
the question: what happens when an
operator, or group of operators, has
3300 3400 3500 3600 3700 3800 3900 4200 only deployed subsets of the allocated
MHz band?
s Fig. 2 Use of the “global” n77 and n78 mid-band spectrum varies by country. Consider the case of n77, which
covers 3.3 to 4.2 GHz. In theory, a sin-
gle n77 solution would support world-
wide use in all regions using this band.
Standards 3GPP n79 4.4 to 5.0 GHz In practice, operators want solutions
Bands 5G
optimized for the subset of spectrum
allocated in their respective regions—
Country China
Allocation Japan in some cases, as narrow as 100 MHz.
If n77 will not work as a global solu-
4000 4100 4200 4300 4400 4500 4600 4700 4800 4900 5000
tion, how about n78, which has a nar-
rower 3.3 to 3.8 GHz allocation? Think
MHz
again. So far, only a few operators
s Fig. 3 China and Japan plan to use small and different portions of the n79 band. intend to deploy the 3.3 to 3.4 GHz
portion of either n77 or n78. Should
narrow subsets of these bands (see Figures 2 and 3). a handset manufacturer be required to enable opera-
Second, FR2 mmWave allocations are following a similar tion in a frequency range that is not even deployed?
pattern, multiplying the problem. Third, many operators Operators vying for the best operation certainly will not
will initially deploy Non-Standalone (NSA) 5G, which in- demand it.
troduces complex, hard-to-manage interactions between Implementing a more regional solution could deliver
5G and regional LTE bands. performance benefits, largely because handset manu-
Those with long memories may remember the dawn facturers can tailor filtering and optimize power and low
of the LTE era, when bands 1 and 7 were heralded as noise amplifier tuning for subsets of the bands. For ex-
global. Unfortunately, both were only adopted in some ample, at initial launch, the majority, if not all, of the front-
regions, not in others. Other bands that have been end modules for n77 will use a non-acoustic filter, which
considered global candidates, such as band 41, were provides good performance for the very wide 900 MHz of
deployed with regional allocation differences. For ex- spectrum—much wider than any LTE band today. When a
ample, the U.S. allocated the full available bandwidth, much narrower subset of n78 is used, say 400 MHz, a bulk
yet China chose only a narrow subset and, in practice, acoustic wave (BAW) filter with steep filter skirts offers
usage was even narrower, since China Mobile was only better performance, providing improved rejection of out-
one of the three operators in China to fully deploy the of-band frequencies and lower insertion loss at the band
allocated frequency range. Only now, many years after edges. This is one example of the tradeoff for handset
launch, is a more unified band 41 allocation structure manufacturers. Focusing on the regional solution would
being considered, with the aim of deploying 5G NR in improve performance for a few targeted mobile opera-
the refarmed band n41. tors, while losing the capability for true global roaming—
The same fragmentation is occurring with the new or at least reducing the number of SKUs to service the
“global” ultra-high bands n77, n78 and n79 and for the global 5G market.
same reason. Little has changed in the way countries A similar dilemma exists for band n79 (4.4 to 5 GHz).
and regions allocate spectrum or auction it to mobile China favors the 4.8 to 4.9 GHz portion of the band,
operators. while Japan is considering 4.5 to 4.6 GHz. A solution
that supports the entire band would work in both coun-
HANDSET WOES tries, but it would not be optimized for either of the nar-
The resulting differences in regional allocations have rower allocations. If you were an operator in one of those
big implications for handset manufacturers, who must countries, would you choose the global n79 solution or
figure out how to support conflicting desires. Operators the one that provides better performance for the users
generally want handsets optimized for the subset of a in your country? On the other hand, as a manufacturer,
band used in their respective regions. However, handset would you want to build separate SKUs for China and
manufacturers want to sell devices globally, or at least Japan or have a single handset to cover both? These
regionally, supporting the different bandwidths and car- decisions trade off more than performance.
rier aggregation (CA) combinations used in all their tar- Regional allocations of FR2 spectrum make the frag-
get markets. mentation problem even more challenging, with differ-
22
 designed for global use, particularly in
Standards 3GPP
regions where 5G is not yet available
n257 n260
Bands 5G n258 or with operators without n78 spec-
n261 trum. In those locations, band 3 would
EU be a main data path, and the added
Country
U.S. losses would penalize handset perfor-
Allocation China mance without providing any benefit
Japan
Korea to the user.
For some 5G bands, the global pic-
24 26 28 30 32 34 36 38 40 42 44 46 48 ture is even more complex, due to the
GHz need to support SA and NSA opera-
s Fig. 4 Use of the mmWave 5G bands will also be fragmented by country. tion. For example, LTE band 41 is be-
ing refarmed as 5G band n41, which
ences between regions and among the operators within will provide much greater single-
each region (see Figure 4). Handset RF implementa- channel bandwidth than the 60 MHz LTE limit. North
tions for mmWave are perhaps more frequency depen- American handsets will need to support dual-transmit
dent than sub-6 GHz implementations, considering the NSA operation using LTE band 41 together with 5G n41.
antennas. If the phone must support multiple, widely This extra 5G bandwidth comes with a penalty, however.
spaced mmWave frequency ranges, it may require mul- Due to reverse intermodulation products created when
tiple antenna arrays or, at the very least, a more com- simultaneously transmitting on both LTE band 41 and
plex and lossy antenna. This could be a requirement n41, the output power must decrease to meet emissions
even within regions, where multiple mmWave frequency masks, potentially reducing coverage. In China, how-
bands are being allocated. Adding to the challenge, it ever, n41 can be used in wideband SA mode without
is likely that mmWave front-ends will be implemented supporting dual transmission. Using n41 in SA mode will
in multiple placements in the handset, each consuming allow a wider bandwidth UL while maintaining the cov-
precious real estate within an already space-challenged erage that operators have achieved with LTE on band
form factor. This is a difficult proposition considering 41, and it can be supported with a single power ampli-
handset size is approaching the practical limit for main- fier, reducing the size and cost of implementation.
taining portability. As regions move to SA 5G, many of these problems
will become simpler. However, that is not likely to hap-
NSA, SA AND LTE pen soon. I believe 5G will be around for a decade or
Regions vary in their initial approaches to implement- more before SA becomes the predominant implemen-
ing 5G. In many areas of the world, operators are plan- tation across the globe. In the race to global 5G deploy-
ning to accelerate 5G deployment by employing NSA ment, we have to accept a long period of RF complexity
5G. NSA uses an LTE anchor band for control and a caused by the NSA standard.
wider, 5G band to achieve higher data rates. This ap-
WILL CONSUMERS BUY GLOBAL 5G HANDSETS?
proach allows operators to deliver 5G sooner, by lever-
aging their existing LTE networks and not building out Despite all these challenges, is it possible to build
a new 5G core. However, some China operators plan to a global 5G handset? Probably. However, will global
quickly transition from NSA to Standalone (SA) 5G or, in roaming justify the cost and other tradeoffs?
some regions, directly from LTE to SA 5G. SA removes One question is whether consumers are prepared to
the need for an LTE anchor, requiring a full 5G network pay for a more expensive global 5G phone. Whether the
buildout, yet easing the implementation of multi-band handset is regional or global, the cost will increase be-
CA combinations, particularly on the uplink (UL). cause the manufacturer will have to pay 5G technology
Although NSA helps operators deliver 5G speeds license fees, in addition to the current fees for LTE. The
sooner, it introduces considerable RF complexity: requir- substantial increase in RF content to provide global 5G
ing dual LTE and 5G connectivity. In many cases, opera- coverage—especially considering the regional band vari-
tors will add a 5G band to the existing regional CA com- ations—will add cost. It may be hard to justify, even for
binations of multiple FDD LTE bands, with one LTE band the specialized group of high-powered business travelers
serving as the anchor. The NSA specification requires who want data roaming wherever they travel.
the handset to transmit on one or more of these LTE Recall the early days of LTE, when LTE was implement-
bands while receiving on a higher frequency 5G band. ed regionally and GSM was used for roaming. When
This increases the likelihood of interactions between the global LTE handsets emerged, quite often much of the
LTE and 5G bands, with harmonics of the LTE transmit handset LTE content was dormant. Content that adds
frequencies potentially desensing the 5G receiver. cost yet is not used is often short lived, especially when a
To illustrate, consider the NSA combination of 5G market matures from its infancy.
band n78 with LTE anchor bands 3 or 66. Harmonic Even with a regional design, adding a single 5G
frequencies generated during transmission on the LTE band increases the RF content and cost. Additional
band fall in the n78 band, potentially reducing receive components are required for the 5G band and to
sensitivity. To alleviate the possibility, additional filtering manage the signal flow through an increasingly com-
could be applied to the band 3 path, which will increase plex signal path to antennas, due primarily to NSA
the losses on band 3. This presents problems in handsets and the myriad CA combinations operators have re-
23
 To accommodate the additional content, 5G phones
may need to be “plus” devices. Even with regional 5G
handsets, standard form factors may not be plausible at
the outset. A quick look around will tell you that not ev-
eryone likes large form factor phones, which represent
the minority of units sold. We may have to await further
size-shrinking technologies before it is feasible to fit 5G
into svelte 18:9 form factor phones that easily fit in a
pocket.

CONCLUSION
5G is driving the market toward further regionaliza-
tion of handsets. Consumers do not want to pay for RF
s Fig. 5 The complexity of a 5G handset will require larger content they will not use, and manufacturers cannot jus-
form factors.
tify the cost of adding content that will be rarely used.
quested. Antennas are already strained to cover exist- Manufacturers may also find it difficult to squeeze the RF
ing frequencies used for LTE; 5G will make the range content into handsets while maintaining battery life and
even wider. While the challenge can be solved using the other features consumers value. The desire to keep
antenna tuning and antenna-plexers, which maximize handset cost and complexity within reasonable limits
the number of signal connections to each antenna, will lead manufacturers to design regional 5G handsets,
both solutions add content to the handset. similar to the regionalization in many of today’s mid-tier
5G will also mean big phones, at least initially (see LTE smartphones.
Figure 5). The additional RF content requires extra To be very clear, none of these factors will slow
space, especially for mmWave coverage, which will be down the impending 5G rollout. They will just fragment
difficult to fit in today’s handset form factors. Manufac- the market. In short, 5G handsets will be anything but
turers do not want to reduce space allocated for batter- global.n
ies and other features that directly appeal to consumers.

24
SAW/BAW New Market Entrants
Offer New Approaches
With Contributions from: Akoustis Technologies, OnScale, Resonant

Editor’s Note: With the increasing number of cellular bands for 4G/LTE, the mobile RF front-end’s critical
component has shifted from the power amplifier to the filter. Surface acoustic wave (SAW), and, more recently,
bulk acoustic wave (BAW) filter technology has been addressing the challenges in the mobile RF front-end that
currently uses 40+ filters (and growing). This market growth has attracted new market entrants so Microwave
Journal compiled information from three such companies—Akoustis, OnScale and Resonant—offering new
solutions for the SAW/BAW market.

XBAW RF Filter Blazing market is currently served by a duopoly that has his-
Into Higher Frequency torically supplied > 95 percent of the market, where
Spectrum both company’s core material technology is based on
a sputtered poly-crystalline piezo-electric aluminum
Dave Aichele nitride (AlN) deposited by physical vapor deposition
Akoustis Technologies (PVD) techniques.
Huntersville, N.C.

B
SINGLE CRYSTAL RF BAW FILTER TECHNOLOGY
AW RF filters are high performance semiconduc-
tor components primarily used in mobile smart Akoustis Technologies is an emerging new entrant in
phones. They address the stringent size require- the projected $5.8 billion BAW filter market dominated
ment for high levels of integration and provide by mobile RF filters.1 Leveraging a patented BAW res-
superior performance compared to SAW and ceramic onator process (called XBAW) combined with an inte-
filters, therefore improving the battery life and reducing grated design and manufacturing (IDM) business model,
the number of dropped calls to end users. These high Akoustis is blazing new territory and focused on becom-
performance components offer low insertion loss and ing the first commercial supplier of BAW RF filters for
high selectivity required to meet the demanding coexis- applications above 3 GHz.
tence requirement for difficult FDD and high frequency Akoustis has introduced a new approach of utilizing
TDD 4G/LTE and emerging 5G bands. Current multi- high purity, single crystal piezo-electric AlN material in
mode, multiband mid- to high-tier smartphones utilize BAW RF filters (see Figure 1). Epitaxially grown, metal-
> 50 filters and experts foresee that including 5G bands organic chemical vapor phase deposition (MOCVD) sin-
will push that number to > 70 filters. gle crystal AlN has inherently higher crystal quality com-
Solidly mounted resonator (SMR) and film bulk pared to PVD poly-crystal AlN. This improved crystal
acoustic resonator (FBAR) are the two dominant BAW quality has shown improvements in acoustic velocity and
resonator technologies currently utilized in BAW RF piezo-electric mechanical coupling coefficients. In addi-
filters due to their high Q-factor, high operating fre- tion, the thermal conductivity of single crystal AlN is 2×
quency and good power handling. The BAW RF filter higher than poly-crystal AlN that degrades as film thick-
www.mwjournal.com/articles/31149
25
ness decreases, which may result in a constraint on pow- RF BAW FILTER MARKETS
er capability for traditional FBAR resonators, especially Akoustis is the only pure play BAW RF filter company
at higher frequencies. In all BAW technologies shown in targeting the mobile high band 4G/LTE and emerging
Figure 1, the resonance frequency is determined by the 5G applications. This market is by far the largest and
thickness of the material stack and the effective propa- made up of filter competitors engaging mobile phone
gation velocity of the acoustic wave. A higher propaga- OEMs and ODMs, RF front-end (RFFE) module manu-
tion velocity in the AlN piezo-material results in higher facturers (some with captive BAW filter technology) and
operating frequencies for the same thickness. These transceiver manufacturers. The push to higher frequency
three factors; improved acoustic velocity, improved and the wide bandwidth requirement necessary to sup-
piezo-electric coefficients and improved thermal con- port the enhanced Mobile Broadband (eMBB) feature
ductivity enable XBAW RF filters constructed from sin- of 5G will tax existing SAW and poly-crystal BAW filter
gle crystal, epitaxially grown MOCVD-AlN piezo-electric technology. Single crystal RF BAW technology will en-
materials to offer better performance (power handling, able the development of higher performance, wider
insertion loss, bandwidth and skirt steepness) than PVD- bandwidth BAW RF filters for 5G n41, n77, n78 and n79
AlN based BAW RF filters, especially for high frequency bands (or sub bands) operating in 2.6 to 5 GHz spec-
and high-power applications (see Figure 2). trum with bandwidths that range from 200 to 900 MHz.
In June 2017, Akoustis completed the strategic ac- Beyond mobile, there are two additional mar-
quisition of a MEMS fab located in Canandaigua, N.Y. kets that will be well served with access to single crys-
With this acquisition and subsequent consolidation of tal RF BAW technology. Advanced Wi-Fi CPE archi-
all its manufacturing processes, Akoustis now has an tectures including 802.11ac multi-user MIMO (MU-
internal, ISO-9001 certified 122,000 sq. ft. commercial MIMO) are experiencing faster uptake, driving the
wafer-manufacturing capability which includes class demand for smaller components as the complexity within
100/1000 cleanroom facility, tooled for 150 mm diam- Wi-Fi infrastructure devices is increasing. This trend is
eter wafers and an operations team to conduct research, expected to continue, especially as 802.11ax is finalized
development and production of its XBAW RF Filters. In and implemented in next generation tri-band routers that
addition, Akoustis is in the process of transitioning DoD operate at 2.4, 5.2 and 5.6 GHz, simultaneously. Ultra-
Trusted Foundry accreditation for MEMs wafer process- small passband 5.2 GHz BAW RF filters provide low 1.2
ing, packaging and assembly, enabling Akoustis to be a dB typical insertion loss over 160 MHz covering U-NII-1
supplier for DoD programs requiring specialized filters and U-NII-2A bands with typical 52 dB attenuation across
and Trusted Foundry certification. 345 MHz to meet the stringent rejection requirements en-
abling coexistence with U-NII-2C and
U-NII-3 bands (see Figure 3). Incum-
Poly Solidly-Mounted Poly Film Bulk Acoustic Single Crystal Bulk Acoustic
bent Dielectric Resonator (DR) filters
Resonator (SMR) Resonator (FBAR) Resonator (XBAW) are 23× larger and require shield cans
Top Electrode Top Electrode
Top Electrode
to mitigate interference issues degrad-
Polycrystal Piezoelectric Layer Polycrystal Piezoelectric Layer
Single Crystal
ing isolation performance.
Bottom Electrode Bottom Electrode Piezoelectric Layer The infrastructure market is looking
Bragg Reflector Air Cavity
Bottom Electrode at full dimension-MIMO or Massive
Air Cavity MIMO architectures which use large
antenna array each with its own trans-
Substrate Substrate Substrate
ceiver configuration to offer much
higher spectral efficiency. These new
s Fig. 1 Cross section images of BAW resonators. basestation systems will probably be

11.6
11.4
Admittance Theta (º)

Single
Output Power (W)

Single Crystal
Crystal 11.2
Vac (km/sec)

Piezo
Piezo
Poly 11.0
Piezo Higher
10.8
10.6
3.7 dB (2.3x) Higher Poly-Piezo
4.7 4.8 4.9 5.0 5.1 5.2 10.4
Input Power (W) Frequency (GHz) 10.2
4K 6K 8K 10K 12K 14K
Thickness (A)

High-Power Handling High Mechanical Coupling High Sound Velocity

High-Power, Wide Bandwidth RF Filters Operating at High Frequency

s Fig. 2 Three performance features of single crystal piezo-electric material.

26
 0
structure applications. Beyond these largest markets, Ak-
oustis is eyeing additional markets such as automotive C-
–10 V2X (or DSRC) and military IF/RF filters for L-, S-, C- and
X-Band phase radar and communication systems.
Insertion Loss (dB)

–20

–30

–40 Enabling Design of

–50
Next-Generation
RF Filters for 5G
–60

–70
Gerry Harvey
5.0 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 OnScale
Frequency (GHz) Cupertino, Calif.

s Fig. 3 5.2 GHz BAW RF filter with typical 1.2 IL and > 50 dB While 4G LTE and LTE-Advanced technologies are
attenuation. still being deployed worldwide, the next generation in
wireless communication promises a paradigm shift in
30 –20 throughput, latency and scalability. By 2025, the emerg-
IL-Pin Sweep to 10 W
ACPR-Pin Sweep to 10 W ing wireless 5G market is expected to reach a total value
25 of $250 billion.2 SAW filters and BAW filters are already
–30 used in 4G devices and will compete for the emerg-
20 ing 5G market. Adoption of 5G will see a significant
ACPR1 (dBc)

increase in the number of filters required in a handset,

IL (dB)

15 –40 with 4G models already employing 40+ filters. This puts

the onus on 5G manufacturers to rapidly innovate new
10
filter designs to capture a share of the growing market.
–50
Such innovations tend to offer a “winner take all” pros-
pect such as the FBAR filter ushering in an entirely new
5
product that captured a large percentage of the 4G/LTE
market segment.
0 –60
30 32 34 36 38 40
Pin (dBm) SAW/BAW DESIGN
To help drive this new level of innovation, OnScale
s Fig. 4 2.6 GHz BAW RF filter—WCDMA adjacent channel has developed a cloud-enabled simulation platform op-
power ratio results. timized for Multiphysics analysis of piezoelectric devices
the primary solution for emerging 5G and an alterna- such as SAWs and BAWs. This approach is being used
tive to traditional macro-cell BTS for 4.5G and 4.9G LTE to reduce cost, risk and time to market for these prod-
networks. FD-MIMO architectures support both FDD ucts.
and TDD bands and offer 1 to 4 W average powers in Optimization of SAW/BAW filters is a challenging task
32T32R to 64T64R configurations operating in the 2 to due to the complexity and size of the devices. OnScale
5 GHz spectrum. These large array systems will need an is well-suited for these kinds of problems, taking a high-
alternative filter technology to enable size/weight reduc- ly efficient Finite Element Analysis (FEA) approach and
tion and high volume, surface mount assembly. Traditional- seamlessly deploying this on the cloud. Figure 5 shows
ly macro-cell style cavity filters are large in size and typically a SAW filter with 100 interdigitated pairs and 20 grat-
require manual assembly so are not ideal for FD-MIMO ing fingers modeled in full 3D. The zoomed-in portion
systems. Poly-crystal BAW RF filters are used in pico- shows the simulated surface velocity at a given time-
and micro-cell BTS but may fall short on power handling step. The entire model can be run in a matter of hours
above 1 W. High-power single crystal technology offers which is a feat even the most powerful legacy solvers on
a potential paradigm shift to the major BTS OEMs de- the market are typically incapable of doing.
veloping 5G FD-MIMO systems. Akoustis has demon- Optimization of this design is achieved using the cloud,
strated BAW RF filter die mounted on standard laminate where hundreds of these models can be simulated simul-
capable of handling > 10 W average power at 2.6 GHz
(see Figure 4). This power handling provides plenty of
power margin headroom for the design of RF BAW fil-
ters that offer smaller form factor, surface mount assem-
bly at semiconductor price levels.

SUMMARY
Akoustis Technologies is a new entrant to the multi-
billion RF filter market and blazing its own path through
material science innovation in single crystal piezo-electric
enabling high performance BAW RF filters in the 3 to 6
GHz spectrum for emerging 5G mobile, Wi-Fi and infra- s Fig. 5 Full-3D model of a SAW filter in resonance.
27
taneously to allow exploration of a design space defined a pentagonal FBAR resonator that has been imported
by the variation of specific parameters. One of these itera- into OnScale from a GDSII file. An image of the instan-
tions reveals a sweet spot where Q is maximized and spurs taneous surface velocity from the simulation is shown in
are minimized in the impedance of the device. Figure 6 Figure 7b.
shows the chosen design’s impedance versus frequency in A major challenge for designers is ensuring that filters
this example. do not support strong lateral resonances that corrupt
passband performance. Resonators with non-parallel
FBAR DESIGN EXAMPLE sides, such as those shown in Figure 8, support weaker
FBAR filters, unlike their surface and bulk silicon lateral resonances than ones with parallel sides. How-
counterparts, use piezoelectric thin films over cavi- ever, optimizing these shapes empirically is expensive
ties with resonant frequencies between 100 MHz and and time consuming. Ideally, an engineer would use full
10 GHz. A range of different shapes and sizes can be 3D simulation for this optimization process, but this is
used depending on the performance requirements, with considered impractical due to the extremely large com-
early designs using square shapes and more advanced putational requirements and time demanded by legacy
designs using pentagons. Figure 7a shows a layout of FEM tools. The cloud method solves this problem, deliv-
ering rapid insights into these complex electro-mechan-
10,000 ical systems and opens entirely new solution spaces for
Reduction in Q vs engineers to explore.
unit cell simulation To demonstrate this capability, a 3D model of a pen-
1,000 tagonal FBAR was constructed and a generic algorithm
Impedance (Ohms)

N Jobs Spawned on Cloud

100

10
Spurs are evident All Results Available in 1 Simulation Timeframe
above and below the
main resonance s Fig. 9 Parallel design study on the cloud.
1
10,000
1.2e+09 1.3e+09 1.4e+09 1.5e+09 1.6e+09 1.7e+09 1.8e+09
Frequency (Hz)

s Fig. 6 Impedance plot of an optimized SAW design in full-3D. 1,000

Impedance (Ohms)

100

10

1
1.80E+09 1.85E+09 1.90E+09 1.95E+09 2.00E+09
Frequency (Hz)
(a)
(a) (b)

s Fig. 7 GDSII import and simulation of a pentagonal FBAR 150

filter.
100

50
Phase (°)

–50

–100

–150
1.80E+09 1.85E+09 1.90E+09 1.95E+09 2.00E+09
Frequency (Hz)
(b)

s Fig. 8 Die photo of an FBAR employing multiple s Fig. 10 Comparison of simple square design (red) and
pentagonal resonators.3 optimized pentagonal design (blue).

28
was used to optimize the design of the
filter to minimize lateral resonances. Package/Module
Chip Layout Layout
Genetic algorithms mimic the process Physical Design
Mask
Generation
Physical Design
of natural selection to guide succes-
sive populations of candidate de- 8 Patents
signs towards a global optimum. Each
Filter Circuit Filter Chip Design Integrated Filter
population of designs was simulated Specifications Design Optimization Design Optimization
in parallel on the cloud, as shown in Optimization Computer Aided Engineering Computer Aided Engineering
Synthesis
Process Synthesis
Figure 9. Parameters Simulation Simulation
Optimization
The model was run for 52 gen- 72 Patents Optimization
Finite Element Model Finite Element Model
erations and a total of 3,640 designs 12 Patents GVR Aquisition 14 Patents
were investigated. It ran for a total
of 68 hours and utilized 8.67 GB of Yield Prediction
Data Analysis
Compliance Matrix Measured
memory. The simulation tool was con- Temperature Behavior & Verification Data
Add External Components
nected to MATLAB’s Global Optimiza- Compare
tion Toolbox, which allowed various Note: Patents issued and pending.
parameters to be tracked during the
run including the current best design. s Fig. 11 ISN schematic, showing process flow from initial design to completed mask.
The optimal designs were found to
have edges angled relative to the substrate edges to S21:Tx-Ant IL and S32:Ant-Rx IL
avoid strong reflections, whereas the worst design had 0
three edges close to parallel with the substrate causing
increased lateral mode activity.
The results can be seen in Figure 10, where the best 10
pentagonal design shows a significant reduction in rip-
ple when compared to the square device, which was the
starting point for the exercise. It is important to note
20
that each of the 3,640 designs were simulated in full 3D,
a study that would take a legacy solver nearly a year to
complete on the same computing resources. The results
Insertion Loss (dB)

provided are indicative of the type of design improve- 30

ments that can be achieved with cloud based CAE using
OnScale’s solvers.
40
SUMMARY
Despite the lack of standards, 5G is promising faster
data rates for mobile phones and will be an enabler for
50
autonomous vehicles and the IoT. The move from 4G to
5G represents orders of magnitude higher data rates at
frequency bands beyond 3 GHz. However, legacy CAE
tools are incapable of performing complete 3D design 60
studies, which are a critical step in optimizing the design
and improving the time to market for these highly com-
plex structures. OnScale’s cloud solvers open the possi- 70
bility of doing this analysis in parallel, reducing prototyp- 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000 2050
ing costs and speeding time to market. Frequency (MHz)

s Fig. 12 Measured (blue trace) and modeled (green trace)

duplexer performance.
Infinite Synthesized Networks
Deliver RF Filter 5G devices will exist in a mobile device environment
Design Tools for 5G that includes more complexity, more components (par-
Bob Hammond ticularly filters), more performance demands, smaller
Resonant
size and lower cost components, plus dual connectivity
Santa Barbara, Calif.
between cellular and Wi-Fi networks. More bandwidth
will be needed, which will require higher frequency
Where 4G/LTE was a single specification for high components, more carrier aggregation (CA), more com-
speed mobile device services, 5G is a family of tech- plex MIMO antennas, new and adaptable waveforms
nologies designed to serve different use cases ranging and improved interference mitigation.
from ultra-broadband fixed wireless to low data-rate IoT 5G RFFE designs for all wireless-enabled products will
services. The transition to this new network technology be driven by cost, power efficiency and available space
will result in dramatic increases in filter and RFFE com- within the mobile device. The requirements for 5G filters
plexity. will include complex multiplexing, increasing integration,
29
 more filters and the capability to handle
SOI Duplexer SOI much higher frequencies than are cur-
HB PAMiD Switch Bank Switch
Module B7 rently in use.
B30
PA Load Line
Match B40
RESONANT INFINITE SYNTHESIZED
NETWORKS
B41
TDD RX Port To address these needs, Resonant
RFFE MPI has developed a comprehensive filter
RX Ports
Electronic Design Automation (EDA)
MB PAMiD
SOI SOI
Quadplexers Switch
platform called Infinite Synthesized Net-
Switch
Module B1 works (ISN). Resonant’s ISN platform
B3 brings together the following elements:
Load Line
PA
Match Triplexer • Modern filter theory.
B25 • Finite element modeling, both
RFFE MPI
B66 electro-magnetic and acoustic.
RX Ports
• Novel optimization algorithms.
• Ecosystem of foundry and
LB PAMiD
SOI Duplexer SOI packaging/back-end partners.
Switch Bank Switch
Module B8 ISN was initially focused on design-
ing acoustic wave filters, which are a key
B20
design block for the RFFE. ISN is spe-
B26 cifically intended to solve many of the
Load Line B12 5G challenges that will face design en-
PA
Match gineers: speed, flexibility and tools that
B13
drive down system cost. As of August
B28A Antenna 2018, more than 10 companies have
Diversity
4G Transceiver

B28B Switch committed to produce more than 60

RFFE MPI devices using ISN.
RX Ports Figure 11 is a schematic that shows
the design-to-mask flow through the ISN
SOI
Switch
RX Filter SOI process. Testing has proven that ISN’s
HB RX Bank Switch
Diversity B7 models are highly accurate and reflect
Module
physical details of the filter structures,
B30 matching the measured performance of
LNA
B40 manufactured filters, not only in loss and
isolation but also in power handling and
B41
RFFE MPI linearity. Thus, ISN is a capable platform
for quickly, efficiently and cost-effective-
SOI RX Filter SOI
MB RX Switch Bank Switch
ly scaling filter design to meet emerging
Diversity B1 5G demand.
Module
Traditional acoustic wave filter design
B3 Triplexer uses a ladder structure and empirical
LNA
B25 models (linked to a particular fab manu-
facturer). This typically results in an itera-
B66
RFFE MPI tive approach to filter development that
LB RX SOI RX Filter SOI involves multiple foundry runs and can
Diversity Switch
B8
Bank Switch take months or more. The ISN platform
Module
enables filter design teams to create
B20 novel filter structures that outperform
B26 traditional filter designs, in a smaller
footprint and using lower-cost technolo-
B12
LNA gies. Figure 12 shows how closely ISN-
B13 modeled performance tracks the actual
B28A
data measured on a Band 3 duplexer.
ISN’s grounding in fundamental ma-
RFFE MPI
B28B terials physics, while optimizing for
high-volume design screening, enables
designs that are unconstrained by tradi-
Wi-Fi/BT/GPS tional acoustic wave filter design tech-
niques. Consequently, a designer using
ISN can create multiplexers, wide pass-
s Fig. 13 Current state-of-the-art RF front-end architecture. bands and high-power performance,
30
 5G FILTER REQUIREMENTS
Total
The growth in the number of filters,
4.6 dB 0.55 dB 0.45 dB 0.75 dB 1.8 dB 0.25 dB 0.65 dB 0.15 dB 0.25 dB
and the ever more demanding perfor-
mance requirements, make RF filtering
PA MN MN the critical pain point of the RFFE. The
TX Traces Traces basic requirements for a 5G filter includes
complex multiplexing driven by CA and
SPnT RX increasing integration to maintain high
Antenna Load
Switch SPnT performance of the RFFE. Maximizing
Antenna
Duplexer
Switch PA efficiency on the uplink, and receiver
sensitivity on the downlink, will require
s Fig. 14 TX path component line-up with estimated losses. optimization of the entire RF chain. As
and predict manufacturing yields as well, before a de- complexity increases, it will be crucial to
sign is committed to mass production. understand the RF chain and any interactions between
Thousands of designs can be developed simultane- elements.
ously and screened to maximize the ultimate perfor- Isolation, loss and power handling requirements
mance of the device. Leveraging the expertise of filter continue to create new performance challenges. Filters
design engineers for an increasing number of more in the RF chain are a major contributor to loss, which
complex designs can be achieved using ISN. is critical for total Tx efficiency (and ultimately for the
current draw for the PA and battery life), and the to-
IMPLICATIONS FOR THE 5G RF FRONT-END tal noise figure in the Rx path (and ultimately for the
ISN can be used to develop RF filters for 4G/LTE and SNR and the data rate). Figure 14 shows the estimated
other wireless networks, but it is especially impactful for losses from each component in the Tx path.
5G designs that need the high performance, small size LTE, which is optimized for high speed data, de-
and complex passband design benefits of the design manded significantly higher power than 3G protocols
tool. such as CDMA. And as such, the requirements for isola-
The current state-of-the-art for a 4G/LTE mobile smart- tion and minimizing leakage into the Rx path, and vice
phone RFFE separates the frequency spectrum into low- versa, grew. This will only be further exacerbated by
band (698 to 960 MHz), mid-band (1710 to 2200 MHz) high-power user equipment (HPUE), which uses more
and high-band (2400 to 3800 MHz) frequencies, which Tx power for improved cell edge coverage. In addition,
isolates the RF components, minimizes cross-talk and power durability of progressively smaller filters becomes
optimizes the entire power amplifier-filter-switch path a major concern.
(see Figure 13). Although integration of components is For 5G, frequencies greater than 6 GHz will require
logical, the increasing complexity of 5G limits the num- different filter technology than the current acoustic wave
ber of manufacturers that have the expertise to execute filters used in mobile devices. Significant advances will
on such a complex RF sub-system. be needed to reduce size and cost. The 5G RFFE for
5G RFFEs for all wireless-enabled products will be mobile broadband will be extremely complex and that
driven by cost, power efficiency and available space the goal for filter design will be to both simplify the de-
within the unit. So they will need to be small, highly ef- sign process and the RFFE itself.
ficient and able to be manufactured in large quantities Innovations that enable 5G RFFEs will need to
to meet fast-growing global demand. To commercial- include a low-loss triplexer (to minimize the num-
ize affordable custom parts for IoT devices in particular, ber of antennas), multi-mode, multi-band PAs and
RFFEs will need to be designed with a minimum number multi-band filters (to reduce the number of filters and
of components and manufacturing volumes will have to switches), all of which will need to be optimized as
increase dramatically from current levels to reduce per- a complete system to reduce matching components.
unit cost. In the current environment, most IoT devices
are being built with low-cost parts originally developed SUMMARY
for high-volume mobile phone production. With RF complexity expected to grow significantly in
As we move toward 5G, the complexity of the RFFE 5G devices, the time is right for a filter design tool that
continues to increase. For instance, in addition to the can design better, more complex components in a time
main antenna path modules, diversity antennas provide and capital efficient way. ISN delivers on this need with
both link robustness and increased downlink data rates. highly accurate, highly integrated and highly manufac-
Designers are increasingly using receive diversity mod- turable filters with complex features.n
ules to process the diversity path, comprised of receive
(Rx) filters and switches (and increasingly incorporating References
1. “RF Front-Ends for Mobile Devices 2018,” Mobile Experts
LNAs). Wireless carriers demanding higher 5G data Inc., 2018.
rates will drive more carrier aggregation, creating more 2. “5G Wireless Market Worth $250 Billion by 2025: $6 Billion
potential interference. Consequently, the onus on RFFE Spend Forecast on R&D for 2015–2020,” PR Newswire, March
designers moving forward will be to reduce complexity, 2016.
reduce cost, while at the same time improving perfor- 3. R. Ruby, “A Decade of FBAR Success and What Is Needed for
Another Successful Decade,” 2011 Symposium on Piezoelec-
mance.
tricity, Acoustic Waves and Device Applications (SPAWDA),
2011, pp. 365–369.
31
WHITEPAPER
Reduce Cost and Complexity in 5G mmWave
systems with Surface Mount Solderable Filter
Components

Introduction The question then becomes that of the availability of

the necessary devices in surface mount packaging. Mass
Frequencies in the mmWave spectrum will play a key
market consumer wireless products have pushed the
role in 5G communications. RF technology that was
development of low cost packaging technologies suitable
developed around existing mmWave applications has
for RF frequencies. With the growing market for SMD
evolved to encompass the needs of 5G wireless access.
RF packages there was a need for size reduction that
Components for such systems need to be selected for
reduced the impact of package parasitics, enabling an
performance and cost – commercial systems are subject
increase in the maximum operating frequency of SMD
to intense price pressure and so both the purchase cost
packaged ICs. In Automotive Applications for example
and the implementation cost of a component become
mmWave frequencies find application in radar based
important factors in selecting devices for a new design.
Advanced Driver Assistance Systems (ADAS), helping
SMD packaging to control cost drivers control vehicles and to assist in automated
functions. These systems often use both short range
The price of the devices themselves are only one part of (at 24 GHz) and long range (at 77 GHz) radar to scan
a designs overall cost. The ability of the component to the environment around the car. The transceivers are
fit into standard assembly processes such as SMD lead packaged in SMD packages to allow the use of standard
free pick and place manufacturing is an important factor production equipment in manufacturing of these
to consider. The Hybrid approach of combining surface systems.
mount with chip and wire assembly could prove to be
a cost driver, since the assembly needs to take place in 5G mmWave Radio Architectures
clean room environments and can result in performance
When it comes to addressing the cost effective assembly
variation from assembly that requires follow up tuning.
of 5G mmWave systems a good place to start would be
By choosing all SMD packaging the manufacturers of
to consider the kinds radio architectures that are being
millimeter wave systems can use standard assembly lines,
deployed and then ask ourselves what device packaging
and will avoid the need for expensive die attachment and
looks like for the major block diagram components.
wire bonding tools in clean room environments.
The standard choices for implementing Full Duplex

WP18TF04 09/20/18 www.knowlescapacitors.com

knowlescapacitors.com 1
32
 WHITEPAPER

communications that can influence radio architecture

has been between Frequency Domain Duplex (FDD)
and Time Domain Duplex (TDD). For 5G systems Transmit

operating < 6GHz both FDD and TDD each have Up-converter PA
Bandpass
T/R Switch Filter

their own benefits. At mmWave frequencies though

Baseband

TDD is generally recognized as the preferred approach

because: 1, Both Transmit and Receive signals are at the Down-converter LNA

same frequency, making adjustments to propagation Receive

and fading characteristics somewhat simpler. 2, TDD

FIGURE 2. Alternative Simplified TDD Architecture
provides improved utilization of wider bandwidths and
On the Receive side in figure 1 the signal at the output
3, Technologies like massive MIMO that are seen as
of the antenna is routed through the T/R switch to a low
essential to 5G are easier to implement with TDD.
noise amplifier (LNA) that has enough gain to achieve
Time Domain Duplex Architectures the target noise figure. The LNA then drives a bandpass
image filter, which is designed to remove received noise
An approach to implementing TDD is with a T/R
that has been amplified by the LNA. From there the RF
switch close a common antenna that is used for both
signal passed through a down-conversion stage and into
Transmit and Receive signals. The T/R switch alternates
the baseband portion of the system for conversion into
between the radios Transmit and Receive modes, routing
digital signals for processing.
signals between each path as necessary. Figures 1 and 2
illustrate a simplified view of a TDD radios architecture On the transmit side of figure 1 the baseband signals
drive an up-conversion stage, which in turn drives a
power amplifier (PA) designed to reach the link budget
goal. The PA drives a bandpass filter, which is designed
Transmit Bandpass
Filter
to suppress broadband noise from the PA and out of
Up-converter PA
band image and carrier feed through products.
T/R Switch

Baseband
Bandpass Filters are necessary on both legs of the journey to
Filter

suppress interference between different systems. Placing

Down-converter LNA

filters in the transmit path to improve out of band

emission reduces adjacent channel interference to other
Receive

systems from the transmitter. Placing filters in the receive

FIGURE 1. Simplified TDD Architecture path improves the receivers adjacent channel rejection of

WP18TF04 09/20/18 www.knowlescapacitors.com 2

33
 WHITEPAPER

interference from other systems. take multiple copies of the simplified block diagrams
in figures 1 and 2 and multiply them across an antenna
Where the filters are placed depends on the particulars of
array, with the addition of the ability to adjust the phase
the system. Figure 1 shows filters placed in line with the
and amplitude of the transmit and receive paths. How
gain stages, where figure 2 places them between the T/R
the phase and amplitude shift is implemented and where
switch and the antenna element.
(is it done in the digital domain, or in the RF, or some
MIMO and Beamforming combination of both) is the subject of some debate.
But in general there seems to be a trend, for mmWave
Earlier we mentioned MIMO (multiple-input and systems at least, towards Hybrid Beamforming, where
multiple-output) as an essential piece of 5G technology. the work of adjusting individual channels is split
MIMO refers to the technique of using more than one between the digital and analog domains.
antenna to send and receive data. MIMO delivers an
increase in channel capacity through Diversity Gains, in Hybrid Beamforming
which multiple Transmitters and or multiple Receivers
In hybrid beamforming a combination of digital and
exchange the same data, increasing the overall signal
analog beamforming is used. N digital paths is split into
to noise ratio (SNR) of the system, and Multiplexing
M RF paths, driving a total of N x M antenna elements.
Gains, in which data can be split across multiple
This architecture combines multiple antenna array
Transmitters and Receivers and then re-combined on the
elements together into a subarray panel. The number of
receive side.
elements in each subarray is selected to ensure that the
A related but distinct concept is that of Beamforming, performance is met while minimizing the complexity of
which is adjusting the radiation pattern of an antenna the design.
array. The technique is well established in Phased Array
In figure 3 some elements from figure 2 are replicated
radar systems where elements of an antenna array are
in the front ends and a beamforming stage is added
arranged, and phase and amplitude of the signal at
between the front ends and up/down conversion blocks.
each element is controlled, in such a way that signals
at particular angles undergo constructive interference The antenna array is split up into subarrays, and each
while in other directions they experience destructive subarray is fed by a set of front ends and beamformers.
interference. This allows a transmitter to ‘point’ signals In this example the design is aiming for one baseband
in a given direction and also for a receiver to ‘listen’ in a path for every 16 active antenna elements, although
given direction also. this ratio varies depending on the intended application
for the beamforming array. We start with a total
Radio systems that utilize beam forming essentially

WP18TF04 09/20/18 www.knowlescapacitors.com 3

34
 WHITEPAPER

Splitter/Combiner (1:4)
Splitter/Combiner (1:4)
Digital Processing x64

Splitter/Combiner (1:4)
Splitter/Combiner (1:4)
Splitter/Combiner (1:4)

x4 x4
x4 x4
x4 x4 x4
Up-converter x4 x4

Down-converter

x4

4 Data Streams

4 RF Beamformers 64 Front Ends x4

4 Antenna Subarrays, Each with 64 Elements

FIGURE 3. Simplified Hybrid Beamforming Architecture

of 4 baseband paths. Each baseband path drives 4 To produce such systems in a standard assembly
beamformers. Each beamformer splits the RF signal 4 processes such as SMD lead free pick and place
ways to drive 4 front end modules, (FEM) and each manufacturing hinges on the availability of key building
FEM feeds an antenna element inside a subarray. This blocks in surface mount packaging. Looking at the key
gives us 64 FEMs per subarray. There are 4 subarrays components of our example architecture (and including
so we have a total of 256 antennas. The panels are the filters that we know will be necessary) we arrive at
dual polarized to the entire array consists of 512 active the list in Table 1.
elements.
One solution for ease of manufacture would be to have
It should be noted that in figure 3 we have left out all of these blocks available in one on chip solution.
the necessary RF bandpass filters. These are likely to This would not be practical however for several
be implemented close to the gain stages, so possibly reasons, including the need by designers to adjust a
between each FEM and its associated antenna element, the architecture to suit the application, the dominance
and/or between the beamformer and the FEM it is of different semiconductor technologies in different
driving, but they may also be necessary between the first blocks and the likely need for the filtering components
splitter/combiner and the up/down conversion blocks. to be implemented off-chip, since on chip solutions
Location of the RF bandpass filters depends on the cannot deliver the necessary performance in real world
interference constraints that a particular design is facing. applications.

SMD for Beamforming Architectures Usually a subset of these components are available as in
integrated modules. So for example mmWave surface

WP18TF04 09/20/18 www.knowlescapacitors.com 4

35
 WHITEPAPER

TABLE 1. KEY COMPONENTS IN THE HYBRID BEAMFORMING ARCHITECTURE AND THEIR AVAILABILITY IN SMT PACKAGES.

Key Component Available at mmWave frequencies? Available in SMT packages?

ADC/DAC na YES
UP/Down Converters and or
YES YES
Modulator/Demodulators
Splitter/Combiners YES YES
Phase Shifters YES YES
LNA YES YES
Power Amplifier YES YES
Filters YES YES
Switches YES YES
Antennas YES YES

mount modulator/demodulators, RF beamformers and shielding:

RF front ends are available on the market today. And
Printed wire board (PWB) covers are one solution
where components need to be stand alone, mmWave
offered by Knowles Precision Devices within the
SMD packaged devices can be sourced.
Dielectric Laboratories (DLI) brand. This style of cover
At Knowles Precision Devices our high performance offers excellent RF shielding for solder surface mount
Dielectric Laboratories (DLI) brand microstrip filters are applications. Additionally, PWB covered components
provided with metallization schemes compatible for both are extremely resistant to high shock and vibration
chip and wire filters, and solder-surface mount filters. environments. These covers attached using epoxy; the
For ease of manufacturing we would recommend using cured assemblies offer a small and sturdy surface mount
the SMD approach but can support hybrid if that is package that can integrate multiple filters in one pc.
your requirement. The overall height of the package is typically 0.1 inch
(2.54mm). However above 10GHz discuss with our
Exceptional performance demands rigorous engineering,
engineering team - performance at higher frequencies
both of the component and of its interaction with the
may be limited.
system. The design of a surface mount filter’s shielding
is a crucial element for achieving laboratory-grade A second option for shielding is the attachment of an
performance outside of the laboratory and assuring integral metal cover to the filter. Sheet metal covers
smooth integration with the system. Shielding protects are compatible with both solder surface mount and
the filter from interference and creates a precisely chip and wire filter applications. Typically, this style
controlled micro-environment for optimal performance. cover is grounded/attached along the perimeter of the
There are three packaging options available for RF part, creating a strong bond and improving overall

WP18TF04 09/20/18 www.knowlescapacitors.com 5

36
 WHITEPAPER

filter isolation. Covers can be recessed to expose the only one part of a designs overall cost, since the ability to
I/O contact pad for chip and wire filters to allow wire- bring a design into the world in a cost effective manner
bonding. The overall assembly height can vary from hinges on the availability of standard manufacturing
0.070 to 0.10 inch (1.78 to 2.54mm). processes.

The third option leaves packaging up to the customer. Mixing technologies, as in the Hybrid approach where
Either the next level of assembly provides the RF chip and wire and surface mount techniques are
shielding for the filter or the customer can have their combined, necessitates an increase in manufacturing
own cover integrated. complexity that can be a cost driver.

The Knowles Precision Devices DLI engineering team Where the key active components in systems such
can provide recommendations for housing dimensions, as the ones described in this article can be sourced
leveraging years of expertise to ensure successful design either as surface mount packaged components or as
integration. If the customer provides their own shielding stand-alone SMD devices, taking a close look at the
for the filter, it is very important that our engineering packaging technology available to implement high
team knows the channel width and cover height that performance filters in a design can save considerable
will enclose the device. These dimensions will be taken cost and complexity when it comes to manufacturing
into account during design and test to ensure that the a new design. The performance repeatability inherent
part will work in its next level of assembly. Where the in both the filter technology itself and the way in
customer is assembling their own cover, the tolerance which the manfacturing approach impacts the overall
of placement of this shield can affect overall filtering circuit repeatability are both keys to this reduction in
performance and should be considered. complexity and cost.

Knowles Precision Devices (DLI brand), provides

high performance surface mount solderable filters in a
reduced footprint compared to filters implemented on
substrates like Alumina and with precision engineered
package shielding that ensures the devices continue to
perform as specified once integrated into a system. Filter
packaging is compatible with many conformal coating
processes as well for robust board assembly.

As mentioned earlier in the article a components price is

WP18TF04 09/20/18 www.knowlescapacitors.com 6

37
We Help the World Stay
Connected.
Providing Millimeter Wave Solutions for
5G Including Small Footprint Surface
Mountable (SMD) Filters.

Dimensions:
0.275" x 0.080" x 0.075"

Home Button
Dimensions:
0.393"

5G sample kits now available for n257,

n258, n260 and n261 frequency bands.

knowlescapacitors.com