Basic Operations on DSP Processor

We have seen the mathematical analysis and various digital systems with digital signals. However, they are of no use if they can't be used in the real world. Since, using a PC for real-time applications is cumbersome. DSP provides a perfect solution to this problem. It is capable of handling large mathematical operations easily while using real-time signals for the manipulations of the data. Thus, DSP can be interfaced with various real-time systems to form an integrated smart system. In this experiment, we acquainted with basic operations of the DSP processor C2000 and carried out siple mathematical operations such as addition, subtraction, bit manipulation,etc.

IEEE Paper Report: Automatic Speech Recognition using Correlation Analysis

This IEEE Paper is on matching the patterns of two audio files to recognize speech. This paper uses FFT and correlation to match the two speech files. As we know, correlation is used to determine the similarity between the two signals. Thus, by correlating the stored signal with the unknown signal, and finding the correlation coefficient the similarity between the two signals can be found out. If the correlation coefficient is close to unity, their exists a strong relation between the two signals and hence, they can be said to be forming same speech pattern.

Patent Review: Correlating Call Data and Speech Recognition Information in a Telephony

Very often speech recognition telephony applications such as Voice Response Systems (VRS) ask a caller to input alphanumeric information for their process. However, it is very inconvenient if the caller has limited display and input capabilities. Also, if one of the characters go wrong, complete call has to be repeated. Hence, there is a need of replacing this alphanumeric input to voice recognition systems. However, background noise and distortion is a hurdle in this area of development.

Patent No.: US 6,931,105 B2

FIR Filter Design using Windowing Method

Previous two blogs demonstrated the design of Digital IIR Filters. However, IIR filters are not stable due their recursive nature. Also, they have non-linear phase response hence, the output of the filter appears to be distorted version of the input.
To eliminate these issues FIR filters are used. They have Linear Phase response and hence, output of the filter is the phase shifted input signal. Also, since they don't have any feedback, the filters are very stable. However, this requires large memory as compared to IIR filters.
One of the method to design FIR filters is using Windowing Method. Other method is using FSM which will be shown in the next post. In this method, a particular window function is chosen on the basis of the attenuation required in the stop band. Using this window function, the coefficients of the filter is found and thus  filter is designed.

Butterworth Filter Design

In the previous blogs, we saw the implementation of various mathematical digital signal processing operations that can be performed on the input signal.
However, the main essence of a processor is to implement a digital filter which can emulate similar characteristics as that  of an analog filter.
Here, to simulate the behavior of digital filters, Scilab Software is used. Scilab provides an easy interface for coding and can also generate frequency response curves using in-built functions.
In this blog, a digital Butterworth Filter is implemented. A butterworth filter is a type of a signal processing filter designed to have a maximally flat magnitude response. Initially, the parameters of the required filter such  as Passband Attenuation, Passband Frequency, Stopband Attenuation and Stopband Frequency are inputted by the user. Based on these parameters, a digital butterworth filter (LPF or HPF) was designed using Bi-Linear Transformation(BLT) Method. Once, the digital filter was designed, order and transfer function was displayed and magnitude response of the filter was shown.

FIR Filter Design using FSM

FSM, i.e., Frequency Sampling Method is another method of designing an FIR Filter. FSM for FIR filter design is perhaps the simplest and most direct technique when a desired frequency response has been specified. It involves uniform sampling of the desired frequency response, and performing an inverse DFT to obtain the corresponding FIR response. The results are not optimal, however, because the response generally deviates from what is desired between the samples. When the desired frequency response is undersampled, the resulting impulse response will be time aliased to some extent. 

Chebyshev Filter Design

Previously, we designed a butterworth filter which is maximally flat. However, the order of these filters are considerably higher. This results in requirement of  large number of components and hence the cost of the filter increases. Also, the roll-off of the butterworth filters are not steep. Thus, in applications where frequency response is more important than the constant amplitude across the frequency range and cheaper filters are required, chebyshev filter is used.
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. Here, too, the parameters of the filter are inputted by the user and complete filter is designed. The ripples can be seen in the magnitude response.

OAM and OSM

In previous experiments, FFT was used on short sequences of input. However, in real time, discretized input sequence is very long. This large sequences can't be processed directly. They are first divided into blocks and then thus the real time signal is processed, using DITFFT, block by block. This block of output sequence is then combined together using OAM or OSM and final output is obtained.

Fast Fourier Transform

DFT being inherently slow, a faster,  a better transform, FFT is used. The number of calculations required by FFT is in logarithmic form, hence a processor requires significantly less number of steps to perform a certain set of calculations. Due to this advantage, FFTs are used in signal processing.

Discrete Fourier Transform

The first transform that was practically implemented on the kit was DFT. The input sequence was converted into a sequence using DFT algorithm. Mathematical manipulations is much easier in DFT. However, the number of calculations required are in exponential forms and hence, DFT requires a large amount of computations.

Convolution and Correlation

The very first experiment performed was based in Convolution and Correlation. Convolution gives the output system in accordance with the input whereas correlation gives the similarity between 2 input signals. The developed code was first verified using Linux GCC Compiler. Once the program was built, it was burnt on the board using Code Composer Studio v4. The output was seen using JTAG emulator interface.