Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
8
Studies and Implementation of Sub Band Coder and
Decoder of Speech Signal
Sangita Roy, Dola B. Gupta and P.K. Banerjee
Abstract--- In the last 40 years a number of coding
techniques for analog sources (speech and images) has been
employed. Sub band coding is a kind of transform coding in
which analog speech signal splitting into a number of different
smaller frequency bands. In this paper stress has been given
on 64 KBPS telephone line speech signal. By sub banding data
rate has been reduced to 12.13804 KBPS. It has been a great
achievement from the point of view of data rate reduction
which in turn saves bandwidth as well as spectrum.
Furthermore this scheme provides acceptable probability of
error and quantization noise. In DM bit rate i.e., bandwidth
requirement is less than PCM. But from the point of view of
signal to noise ratio (SNR), PCM is better than DM, in this
dynamic range of SNR at the cost of slightly higher channel
bandwidth. In this paper the PCM, DM have been studied
which are most popular type of coding techniques used
commercially. We have shown that if the SNR output required
is above 30 dB, PCM outperforms DM which subsequently
used in sub band coding. It is interesting to note that 30 dB is
the minimum requirement for communication systems [3].
Keywords--- DM, PCM, SNR, SUBBAND, Probability of
Bit Error
I.
INTRODUCTION
S
UB-BAND coding (SBC) is a kind of transform coding. A
signal is divided into a number of different frequency
bands and encodes each one independently. It enables a data
reduction / compression by discarding information about
frequencies which are masked. The result differs from the
original signal, but if the discarded information is chosen
carefully, the difference will not be noticeable, or more
importantly, objectionable.
II.
a filter bank (sub-band coding), or by a suitable transform
(transform coding), and then encode them using adaptive
PCM. Three basic factors of designing of the coders: 1) the
type of the filter bank or transform, 2) the choice of bit
allocation and noise shaping properties, and 3) the control of
the step-size of the encoders. Short-time analysis/synthesis,
practical realizations of sub band and transform coding are
interpreted within this framework. Spectral estimation, models
of speech production, perception and the “side information”
can be most efficiently represented and utilized in the design
of the coder (particularly the adaptive transform coder) to
control the dynamic bit allocation and quantize step-sizes.
Recent developments and examples of the „Vocoder-driven”
adaptive transform coder for low bit-rate applications is also
discussed in [5].In digital Telecommunication system different
signals are processed with different sampling rates, these
arises significant errors. In ”Sub band Coding of Speech
Signals Using Decimation and Interpolation”- a structure of a
two-channel quadrature mirror filter with low pass filter, high
pass filter, decimators and interpolators, is proposed to
perform sub band coding of speech signals in the digital
domain. The performance of the proposed structure is
compared with the performance of the delta-modulation
encoding systems. The results show that the proposed
structure significantly reduces the error and achieves
considerable performance improvement compared to deltamodulation encoding systems [6].
III.
PERFORMANCE OF PCM OVER DM
If it is assumed that each of the digital words has n binary
digits, there are M= 2n unique code words. If R= bit rate, n =
no. of bits in PCM, f s= sampling rate, M= quantizing level,
B=bandwidth of analog system, then the bit rate is R=nfs and
BPCM > n f s /2 =n B.
PCM signal to noise ratio can be expressed as
LITERATURE SURVEY
The work done in [4] compared SBC and RIQ to
conventional coding techniques. The system shows SNR 2
to 5 dB higher than that of other coders of similar
computational complexity of wideband audio signals. The
basic concept of “Frequency Domain Coding of Speech”
methods is to divide the speech into frequency components by
Sangita Roy, Electronics and Communication Engineering, Narula
Institute of Technology, under West Bengal University of Technology,
Agarpara, Kolkata –700 109, India.
Dola B. Gupta, Electronics and Communication Engineering, Narula
Institute of Technology, under West Bengal University of Technology,
Agarpara, Kolkata –700 109, India.
P.K. Banerjee, Electronics and Tele communication Engineering
Jadavpur University, Garfa Main Road, Jadavpur, Kolkata, West Bengal.
(S/N)PCM= 6.02n + α
(1)
α=0for average, α = 4.77 for peak SNR .This equation is
called the 6-dB rule [2]. The S/N of speech signal for the DM
system is
(2)
(S/N)DM =3f 3s ‹ w 2(t) › / ((1600П) 2 B w 2 p)
‹ W 2(t) › /w2p = average audio power to peak audio power
ratio.
From equations 1 and 2, figures 1 and 2 have been
developed where for α = 4.77 PCM overtakes DM after 23.65
dB and for α = 0 after 30 dB.
ISBN 978-93-82338-06-2 | © 2012 Bonfring
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
9
compative study of SNR for DM and PM for a=0
70
snr1: DM
snr2:PCM
60
snr in dB
50
40
30
20
10
0
0
1
2
3
4
5
6
fs= sampling frequency in HZ
7
8
9
4
x 10
Figure1: Comparative Study of DM and PCM with α=4.77
compative study of SNR for DM and PM for a=0
70
snr1: DM
snr2:PCM
60
snr in dB
50
40
30
20
10
0
0
1
2
3
4
5
6
fs= sampling frequency in HZ
7
8
9
4
x 10
Figure 2: Comparative Study of DM and PCM with α=0
Transmitter (figure 5) consists of one LPF and six BPFS.
IV.
DESIGN PROCEDURE FOR SUB BAND CODING FOR
All BPFS outputs are multiplied by the lowest frequency
SPEECH SIGNAL (A BASIC SBC SCHEME)
component of corresponding bands at the multiplier block.
The Power Spectral Density (PSD) model of an Voice The outputs are PCM and then added by summer. Finally the
Signal considered is shown in figure 4, and Voice signal has outputs are summed and put into channel.
been restricted to 3.5 KHz only.
In Figure 4, frequency axis is divided into number of
subbands (say 0-f1, f1-f2, f2-f3, f3-f4, etc.). The frequency band (0f1 ) is base band signal whereas ( f1-f2) ,( f2-f3 ),( f3-f4 ) ,etc are
band pass signals. Each band will translated to baseband by
multiplying with lowest frequency component of the said
subband. Here seven subbands have been considered.
ISBN 978-93-82338-06-2 | © 2012 Bonfring
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
10
Figure 3: Power Spectral Densities vs. Frequency of Speech Signal
Figure 4: Power Spectral Densities vs. Frequency of Speech Signal Using Sub Band
Figure 5: Block Diagram of Sub Band Coding Transmitter
At the receiver (Figure 6) signals are decoded by seven components and then passed through BPFs of f2 - f1, f3 - f2 etc.
decoders. Then each signal is passed through LPF of cut-off Then the outputs are summed up to get the replica of the
frequency f1, f2 - f1, f3 - f2 etc . From second to seventh signal original signal.
outputs are multiplied by their respective lowest frequency
ISBN 978-93-82338-06-2 | © 2012 Bonfring
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
11
Figure 6: Block Diagram of Receiver
Figure 7: Staircase Approximation of PSD vs. Frequency of Voice Signal
Data rate from the above signal (Figure 7) is reduced from
Further data rate can be reduced, if multiplied by the
64 KBPS to 19.5 KBPS, according to the considered model probabilities of occurrences using a practical voice signal of
and assumptions.
15 sec. duration (Figure 8).
ISBN 978-93-82338-06-2 | © 2012 Bonfring
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
Figure 8: Original Powers Spectral Density Voice Signal of 15 Sec Duration
.
Figure 9: Probability of Occurrence vs. Frequency of Speech Signal
plotting of frequency vs. probabilities of occurence and power spectral densities of voice signal
40
Probability and power spaectral density
35
30
25
20
15
10
y1: Probability of v/s
y2: Power Spectral Density of v/s
y : Resultant PSD OF V/S
5
0
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
3.5
Figure 10: Resultant PDF and PSD vs. Frequency of Speech Signal
ISBN 978-93-82338-06-2 | © 2012 Bonfring
12
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
13
plotting of frequency vs. Data Rate in KBPS
2.5
r : Data rate
Data Rate in KBPS
2
1.5
1
0.5
0
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
3.5
Figure 11: Data Rate vs. Frequency of Figure 10
plotting of frequency vs. Data Rate in KBPS
14
r : Data rate
cumur:Cumulative Data Rate
12
Data Rate IN KBPS
10
8
6
4
2
0
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
3.5
Figure 12: Cumulative Data Rate as Well As Data Rate
In figure 13 and 14, it have been shown that frequency vs.
From Figure 10, data rate can be further reduced to
quantization error, SNR and another frequency vs. probability
12.13804 KBPS which is much lower than 19.5 KBPS.
of bit error are negligible i.e. within the tolerance limit(10 -5 –
Data rate and cumulative data have been shown in the
10-6).
figure 11 and 12 respectively. Now it is very essential to find
out SNR, quantization noise produced out of sub banding.
ISBN 978-93-82338-06-2 | © 2012 Bonfring
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
plotting of frequency vs. Step Size ,Quantization Error
Step Size,,quanti.erroe in dB ,signal-to-noise ratio
60
qerrordB
snr
50
40
30
20
10
0
-10
-20
-30
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
3.5
Figure 13: Quantization Error, and Signal-To-Noise Ratio
plotting of frequency vs. Probability of bit error
1
Probability of bit error
0.8
Probability of bit error
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
Figure 14: Probability of Bit Error vs. Frequency
V.
PERFORMANCE STUDY
Now parameters - bit rate, probability of bit error of sub
band speech signal under investigation will be compared with
the existing 64 KBPS telephone line.
ISBN 978-93-82338-06-2 | © 2012 Bonfring
3.5
14
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
15
plotting of frequency vs. Data Rate in KBPS
60
Data rate
Cumulative Data Rate
Existing data rate
cumulative existing data rate
Data Rate IN KBPS
50
40
30
20
10
0
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
3.5
Figure 15: Comparison of Sub Band Data Rate and Existing Data Rate
plotting of frequency vs. Probability Error
1
Probability of Error
64
0.8
0.6
Probability of error
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
0
0.5
1
1.5
2
2.5
voice frequency range in KHz
3
3.5
Figure 16: Probability of Bit Error of Sub Band and 64 KBPS Line vs. Frequency
From the above figure it is clear that existing data lines are data rate as well bandwidth savings without losing any
of high data rate, requiring more number of bits and large significant information and probability of bit error is also least
bandwidth as compared to sub band coding which offers very or may be said negligible. PCM requires high bandwidth as
low data rate, less number of bits as well as bandwidth. well as data rate. But PCM and DM have almost same SNR up
Another comparative study can be performed from the point of to 30 dB. After 30dB PCM shows performance wise better
view of bit error. It is clear that both the proposed scheme and results than DM. It should be stated here again that 30 dB is
64 KBPS have acceptable probability of bit error and the minimum criterion for any communication system. Hence
quantization noise.
it can be concluded that PCM proves to be better than DM
which is again outperformed by Sub band Coding. Further, if
more sub bands are used, data rate can be reduced more and
VI.
CONCLUSION AND FUTURE WORK
It is evident from the above discussion that both sub band more accurate approximation of the original voice signal can
coding and existing 64 KBPS line have almost negligible be reconstructed.
probability of bit error but sub band offers lowest data rate,
bandwidth ever possible. Therefore it can be deduced that sub
banding generate all the possible significant footsteps towards
ISBN 978-93-82338-06-2 | © 2012 Bonfring
Proceedings of National Conference on Electronics, Communication and Signal Processing (NCECS 2012), 19th September 2012
VII.
ACKNOWLEDGEMENT
Authors deeply express their sincere thanks to the Head of
the department of ECE for encouraging and allowing them to
carry out the project and related support i.e., Simulation
laboratory whenever they required. Authors took this
opportunity to thank their all faculties who have directly or
indirectly helped their paper. Last but not the least they
express their thanks to family, friends and colleagues for their
cooperation and support.
VIII.
[1]
[2]
[3]
[4]
[5]
[6]
REFERENCES
John G. PROAKIS, Digital Communications, Mc graw-Hill
International Edition, fourth edition
LEON W. COUCH II, Modern Communication Systems Principles and
Applications, Prentice –Hall of India Private Limited,1995
Herbert Taub, Donald L Schilling, Principles of communication
systems, Tata Mcgraw-Hill publishing company limited, second
edition
Yang-Jeng Chen, Robert C. Maher ,Sub-band Coding of Audio using
recursively indexed quantization, Department of Electrical Engineering
and Center for Communication and Information Science, University of
Nebraska-Lincoln
Jose M. Tribolet, Member, IEEE, and Ronald E. Crochiere, Senior
Member, IEEE, Frequency Domain Coding of Speech, IEEE
Transactions on acoustics speech and signal processing, vol.Assp-27,
No. 5, October 1979
Ashra f M. Aziz, Sub band Coding of Speech Signals Using Decimation
and Interpolation, 13th International Conference on AEROSPACE
SCIENCES & AVIATION TECHNOLOGY, ASAT- 13, May 26 – 28,
2009, Military Technical College, Kobry Elkobbah, Cairo, Egypt
ISBN 978-93-82338-06-2 | © 2012 Bonfring
16