Study of a Novel ADM Algorithm with Pre-processing for Performance Improvement

by B.K.Sujatha

M.S. Ramaiah Institute of Technology, BangaloreGuide: Co-Guide:Prof. Dr. P.S. Satyanarayana Prof.Dr. K. N. HaribhatThe Head, Dept of Electronics &Comm Engg The Head, Dept of Telecomm EnggB.M. Sreenivasaih College of Engineering Nagarjuna College of Engg and TechBangalore Bangalore

Topics SPEECH CODINGDISCRIPTION OF EACH CODING TECHNIQUEADAPTIVE DELTA MODULATIONPRE-PROCESSING STEP SIZE ALGORITHMEXISTING STEP SIZE ALGORITHMSSONG ALGORITHMMODIFIED ABATE ALGORITHMPROPOSED ALGORITHM CONCLUSIONBIBLIOGRAPHY

ADVANTAGES AND DIS-ADVANTAGES OF DIGITAL COMMUNICATION

ADVANTAGES:• Less distortion in the received signal.• Simple and less expensive digital circuitry.• Possibility of processing digital signals.• Better received speech quality.• Possibility of transmission of voice,video,data all in digital

form.• Possibility of correction of medium errors.• Encryption/decryption for message security.DIS-ADVANTAGES:• Increased bandwidth.

• Synchronization requirement.

SPEECH CODINGConversion of analog speech signals into digital form

Types of speech coding:

• Pulse Code Modulation• Differential Pulse Code

Modulation(DPCM)• Delta Modulation(DM)• Adaptive Delta Modulation(ADM)

PULSE CODE MODULATION

Steps involved in PCM :• Sampling • Quantizing• Encoding n = log2L

Bandwidth of PCM depends on bit rate, R =

nfs

For no aliasing, fs >= 2 fm

BPCM >= ½ R = ½ nfs

DIFFERENTIAL PULSE CODE MODULATION

• To minimize redundant transmission

• To reduce the bandwidth in comparison with PCM

DELTA MODULATIONONE BIT OR TWO LEVEL VERSION OF DPCM: This one-bit codeword eliminates the need for'

word framing’ at the transmitter & receiver & makes DM systems very attractive for many classes of digital communications.

NOISE IN DM :• Smaller step size causes slope overload distortion.• Larger step size causes granular noise.

DM WAVEFORMS

LIMITATIONS OF DM

m(t)

stair case approximate m(t)

Max slope overload

Slope overload (positive)

^

m(t)

m(t)

Negative slope overload

Slope overload (negative)

^

LIMITATIONS OF DM (contd..)

m(t)

m(t)

Granular noise

(slow varying signal)

LIMITATIONS OF DM (contd..)

^

ADAPTIVE DELTA MODULATION• Improved version of DM by making the

step size of the modulator assume a time varying form.

• Here the step size is adapted to the level of the input signal

ADM WAVEFORMS

Sample speech signal

The sample speech waveform in the illustration is taken from the speech sound “i i i i i” which is shown in Figure. It is one of the waveforms used repeatedly in the simulation that is about 5s long.

Pre-ProcessingA methodology for further improving the

ADM performance by pre-processing the speech signal prior to the adaptation is presented.

The large variations in the speech are removed/smoothened by a suitable pre-processing method, one of which is using an integrator which can smoothen the rapid changes.

At the receiver, the differentiator is followed by a low pass filter(LPF).

m(t)

(2)(1)

(1)Slope overload distorti

n (2)Granu

PRE-PROCESSING OF MESSAGE SIGNAL

m(t)dt smoothes out m(t) , rapid changes may disappear.

t

m(t) (2)

(1)

(1)Slope overload distortion region

(2)Granular noise region

t

Frequency response of Pre-Processor (Integrator) at the transmitter

Frequency response of the Differentiator at the receiver

ENCODER

DECODER

The block diagram of Conventional ADM

ENCODER

DECODER

The block diagram of proposed ADM

STEP-SIZE ALGORITHM:

• In the step-size algorithm, the processor detects the pattern of e(t) where

e(t) = sgn [m(t)-m(t)]• To see if the delta modulator is operating in the

quantization region, in which case e(t) produces an alternating …1010… pattern, or in the slope overload region in which case e(t) produces an all 1’s or all 0’s pattern. These cases are illustrated as shown.

• If ADM senses a 1010 pattern, it decreases the step size, and if it senses …1111… or …0000…, it increases the step size . The manner in which the step size is altered determines the algorithm.

^

Linear delta modulation and the bit pattern produced for each region

t

e(k) 1 01 0 1 0 1 1 1 1 1 1

m(t)

m(t)^

EXISTING STEP SIZE ADAPTATIONS

SONG ALGORITHM• Here, we see that as long as e(k) is of the same sign

as e(k-1), the magnitude of the new step size s(k+1) will exceed the magnitude of the old step size s(k) by so, the ‘min step size’.

• However, if e(k) and e(k-1) differ in sign , the magnitude of s(k+1) will be less than the magnitude of s(k) by the amount so.

• The equation describing the song algorithm is given by

│s(k)│+ so, e(k) = e(k-1)|s(k+1)|= │s(k) │- so , e(k) e(k-1)

MODIFIED ABATE ALGORITHM• The need to maintain voice communications

as long as possible was a key factor in the selection of the modified abate algorithm.

• The equation describing modified abate algorithm is

[|S(k)| + So] e(k); e(k)=e(k-1) and S(k) < 8So

S(k+1)= |S(k)| e(k); e(k)=e(k-1) and S(k) = 8So

So e(k); otherwise

The Proposed step-size adaptation The new proposed technique for the step-size adaptation is

described as  [|S(k)|+S0]e(k); e(k)=e(k-1)

S(k+1)= [β|S(k)|-S0]e(k); e(k)≠e(k-1)

and β| S(k)|> S0

S0e(k) ); e(k)≠e(k-1)

and β| S(k)|< S0

is the adaptation parameter, nearly equal to 1 but, greater than 1.

β 1/()

The Proposed step-size adaptation (cont…)

This adaptation parameter gives a better performance to slope overload

The parameter β takes care of the granular noise as a result of which a better performance is obtained as compared to SONG and modified ABATE algorithms.

Where is taken as 1.1 and S0 as equal to 0.1.

SIMULATIONSNR CALCULATION

{Xn } → samples of original signal (speech signal)

{Xn } → samples of final reconstructed signal

(Xn - Xn ) → error signal

(Xn -X n )2 → squared error signal 

where N is the total sample number of the input.  OR

^

^

^

(a) Performance Comparison of the proposed step-size adaptation algorithm with the SONG and the modified ABATE algorithms.

(b) (b) the same plot of figure.(a) is shown but the input strength is displayed for -7db to -1db.

(a) (b)

(a)Performance Comparison of the proposed ADM with the SONG, modified ABATE and the proposed algorithms. (b) the same plot of figure.(a) is shown but the input strength is displayed for -7db to -1db.

(a) (b)

CONCLUSIONSimulations are carried out for all the schemes. S0 is

taken as 0.1 and Simulations have also confirmed that with the input strength for -7db to -1db on an average a 1.1dB performance gain in the SNR is got for the new step-size adaptation algorithm compared to the SONG and a 1.5dB performance gain compared to the modified ABATE algorithm.

Next, with the proposed methodology(pre-processing) and with the same input strength, on an average there is 1.4dB performance improvement in the SNR for the new step-size adaptation algorithm as compared to the SONG and a 1.7dB compared to the modified ABATE algorithm.

References [1] U N Chong Kwan, Hwang Soo Lee, “A study of the comparative performance of Adaptive Delta

Modulation systems”, IEEE Transaction vol 28, Jan 1980. [2] Donald L, Schilling, Joseph gorodnick and Harold A.Vang, “Voice encoding for the space shuttle using

ADM” IEEE Trans. on Communications, vol. COM-26, no. 11, Nov 1982. [3]. G. S. Tombras and C. A. Karybakas, “

New adaptation algorithm for a two-digit adaptive delta modulation system,” International Journal of Electronics, vol. 68, no.3, pp.343–349, 1990.

[4] M. A. Aldajani and A. H. Sayed, “A stable adaptive structure for delta modulation with improved performance”, Proc. of IEEE International Conference on Acoustics, Speech and Signal processing (ICASSP ’01), vol.4, pp 2621-2624, Salt Lake city, Utah, USA, May 2001.

[5] M. A. Aldajani and A. H. Sayed, “Stability and performance analysis of an adaptive sigma-delta modulator,” IEEE Transactions on Circuits and Systems II, vol. 48, no. 3, pp. 233–244, March 2001

[6] K.Yao, K K Paliwal and S.Nakamura, “Noise adaptive speech recognition with acoustic models trained from noisy speech evaluated on Aurora-2 database”, Proc. Intern. Conf. Spoken Language Processing, Denver, Colorado, USA, pp.2437-2440, Sep.2002.

[7] Ming Yang, “Low bit rate speech coding”, Potentials, IEEE, vol 23, No. 4, pp. 32-36, Oct-Nov. 2004. [8]Gibson J D, “Speech coding methods, standards and applications”, Circuits and Systems Magazine,

IEEE, Vol. 5, No. 4, Pp. 30-49, fourth Quarter 2005. . [9] Hui Dong, Gibson J D, “Structures for SNR scalable speech coding”, IEEE Transactions on Audio,

Speech and Language Processing, vol. 14, No.2, pp.545-557, March 2006 [10] E. A. Prosalentis and G. S. Tombras, “

A 2-bit adaptive delta modulation system with improved performance,” EURASIP Journal on Advances in Signal Processing, vol. 2008, Article 9, 5 pages, 2008.