Learning Experience of Digital Signal Processing and Processors Laboratory

           Real-Time speech recognition system is essential for the hands-free applications, this paper did the hardware implementation of real-time speech recognition system, by two approaches and their performances are evaluated. The first approach uses Mel Filter Banks with Mel Frequency Cepstrum Coefficients (MFCC) as feature input and the second approach uses Cochlear Filter Banks with Zero-crossings (ZC) as feature input for recognition.Recognition systems are implemented on a Texas Instruments TMS320C6713 floating point digital signal processor.The recognition performance is also studied for different combinations of test and training.The procedure is 1. sampling 2.End point detection 3.MFCC/Cochlear 4.SVM classifier 5.Recognized word.According to authors, The proposed real-time speech recognition system can be easily employed for various hands-free applications. Paper- Hardware Implementation of Real-Time Speech Recognition System using TM...

In this Laboratory session, the aim was to perform the basic operations on DSP Processor specifically on Texas Instruments C2000 DSP(Digital Signal Processor) TMS320F28335.  The platform was Code Composer Studio 3 where the code was written in C language.The result various registers was noted down before and after each execution which is the major difference while performing operations on DSP board and the C compiler.  The output of these instructions is seen using the JTAG debugger. DSP Processor allows faster and more efficient execution of this instruction with the help of Multiply and Accumulate Unit (MAC). Basic operations were performed on the C2000 Board such as addition and subtraction.We have also performed the Shift operations on DSP board

frequency Sampling Method is an another way of designing linear phase FIR(Finite Impulse Response) Filter. The process of obtaining the desired frequency response is same as the window function method. The desired frequency response is sampled in the frequency domain,  Discrete Fourier Transform(DFT) of samples was taken. FIR filter response is now obtained by taking Inverse Discrete Fourier Transform(IDFT). The results were verified by comparing the stopband and passband attenuation from the magnitude and phase spectrum which were plotted on the Graphic window of the Scilab.

The aim of the experiment was to design linear phase FIR (Finite Impulse Response) filter using window function.In this experiment, we have used hamming window function to design Low pass filter (LPF) and high-pass fiter(HPF)  Bandpass/band stop filter. We have plotted magnitude as well as phase response. There are  different types of windowing functions Rectangular, Bartlett, Hamming, Hanning, and Blackman.The specifications of As, Ap, stop band frequency, passband frequency, and sampling frequency was taken from the user.It is observed that as the order of the filter increases the number of lobes in frequency response increase.The attenuation depends upon the type of window used.  The phase response of the filter is linearly varying with frequency and no distortion is observed in the output.

   In the last Experiment, we have performed the Butterworth filter. The aim was to design Chebyshev filter, Chebyshev filters are filters which have ripple in their frequency response. Filters having ripple in the passband and monotonic in stopband are called Chebyshev-1 filters. Whereas, filters having ripple in their stopband and monotonic in passband are called Chebyshev-2 filters. The design procedure was again on Scilab.We can conclude from the experiment that,  Poles of the Chebyshev filter lie on ellipse where poles of Butterworth filter lie on the unit circle.Order of Chebyshev filter is less than the order of Butterworth filter and thus realization of Chebyshev filter requires less hardware.  

  A filter is a very important part of the system, so the task was to design the Butterworth filter . This experiment was performed using Scilab which is the open source alternative to MATLAB.We have designed both filters i.e.High Pass Filter and Low Pass Filter using Scilab. Here the parameters like passband attenuation (Ap), stopband attenuation (As), analog passband frequency (Fp), analog stopband frequency (fs) and sampling frequency (Fs).were taken as inputs. Sampling frequency selected almost 4 times higher than the highest frequency parameter,  for LPF Fp<Fs<fc and for HPF Fs<Fp<Fc  Based on these parameters, a digital Butterworth filter (LPF and HPF) was designed using Bi-Linear Transformation(BLT) Method. Once, the digital filter was designed, order and transfer function were displayed and magnitude response of the filter was shown in the Scilab Graphic window.

          The Aim of this experiment was to perform filtering of long data sequence using Overlap Add Method and Overlap Save Method.  Time-domain based convolution require the entire input sequence to be present before commutation. This can pose a problem in real time applications where the data input is long and continuous. Overlap-save method and overlap-add method divide the input sequence into smaller blocks. The filtering of individual blocks which are smaller in length is done by DFT and IDFT and then fitted one after the other. The overlapped portion is discarded in OSM whereas it is added in OAM to get the required output sequence.We wrote the program in C and obtained the convolution for different input signals like N=14,17,18.etc. This resultant is the same as that if the input had been processed via time domain convolution

          This was the great experience of learning digital signal processing, the aim of this experiment was to study mathematical operations such as linear, circular, convolution linear using circular convolution. I developed a function in C program on Linux terminal to find the LC, CC, LC using CC and verified the output of the program using mathematical time domain method.In linear convolution, if both signals are causal then the resultant output signal is also causal.Convolution gives the output of the system, circular convolution gives aliased output.        Let's come to correlation, it has two types auto -correlation and cross-correlation.An autocorrelated signal is an even signal.Autocorrelation of delayed signal is same as the original signal.The application of correlation is the  measurement degree of similarity between two signals, we used Carl's correlation coefficient formula to find the degree of sim...

       This was great learning experience the objective of this experiment was to perform Fast Fourier Transform and to develop a program to perform FFT of N-point signal.Then we did programming in C on Linux terminal and compiled a program to find FFT( N=4,8). Then we did the computation of real and complex multiplication and addition.S peed plays a very vital role and hence FFT algorithms are preferred to obtain DFT of a sequence. Fast Fourier Transform reduces the number of computations required considerably. Thus, improving the efficiency by reducing the time and memory requirements of the system.

 This was the great experience of learning Digital Signal Processing Laboratory, the aim was to perform Discrete Fourier Transform .in this lab session I have developed a function in C program on Linux terminal to perform the DFT of N-point signal where N was taken from the user (4/8). Then I performed the DFT of 4 point signal and 8 point signal and plotted the Magnitude spectrum. DFT gives output in a frequency domain, DFT spectrum is also discrete spectrum and it is approximated.DFT coefficients are defined as W=2*pi*k/N, it gives periodic results.          Also,  I have done some more operation, I increased the length of the input signal by zero padding and then obtained the DFT of the signal after result analysis I found that as the length of signal increases by zero padding frequency spacing decreases, approximation error decreases and Resolution of the frequency spectrum increases.Also, expansion of the signal in time domain gives comp...