Basic operations on DSP Processor

In this Lab session some basic arithmetic and logic operations were performed on  DSP(Digital Signal Processor) TMS320F28335. These operations were performed using assembly instructions defined by the processor's architecture. The output of these instructions is observed using the JTAG debugger. 
      These operations are similar to the ones in a regular processor but the DSP's hardware allows faster and more efficient execution of these instructions in a single instruction cycle. Multiply and add instruction(MLA) also has a single cycle execution due to a dedicated multiplier block inside the DSP. These fast operations differentiate DSPs from other processors. 
       As multiplication and addition are commonly used in signal processing applications, the DSP hardware is used in such applications where faster computation is required. Also these instructions can be executed million times in few seconds.

Paper Review:Correlation based pitch extraction method in speech signal

    This paper proposed a correlation based method using the auto-correlation function . The auto-correlation function is a popular measurement in estimating pitch in time domain. The performance of this method, however, is effected due to the position of dominant harmonics (usually called as the first formant) and the presence of spurious peaks introduced in noisy conditions.

    The aim is to detect the pitch information of noiseless and noisy speech.  The experimental results of computer simulations on male and female voices in white noise perform that the gross pitch errors are lower in proposed method as compared to other related method in different types of signal to noise ratio conditions.

To evaluate the proposed method, natural speech signals spoken by two Japanese female and male speakers are examined. Speech signals are taken from a particular database . The reference file is constructed by computing the pitch frequency every 10 ms using a semi automatic technique based on visual inspection. White Gaussian noise is added to these speech signals before the simulations. Pitch estimation error is calculated as the difference between the reference and estimated pitch frequency. If the estimated pitch for a frame deviates from the reference by > 20%, it is recognized as a gross pitch error (GPE) . Otherwise, a fine pitch error (FPE). For the GPE, proportion of the error was calculated. The possible sources of GPE are pitch doubling, halving and inadequate suppression of formants to affect the estimation. In summary, the proposed method can estimate stable pitch with high accuracy not only in a clean speech but also in heavy noisy conditions.

Patent review : Detecting emotion in voice signals using correlation

The patent emphasises the implementation of a system for detection of emotions in a real time voice signal.

First, a database is provided. This includes statistics of human associations of voice parameters with emotions. Next, a voice signal is received. Minimum one feature is extracted from the voice signal. Then the extracted voice feature is correlated with the voice parameters in the database. An emotion is automatically selected from the database based on the correlated output of the extracted voice feature and the voice parameters and is then output.

     The feature that is extracted includes a maximum value of a fundamental frequency, a standard deviation of the fundamental frequency, a range of the fundamental frequency, a mean of the fundamental frequency, a mean of a bandwidth , a standard deviation of energy, a speaking rate, a slope of the fundamental frequency, a maximum value of the energy, a range of the energy.

       The database includes probabilities of a particular voice feature being associated with an emotion. Preferably, the selection of the emotion from the database includes analyzing the probabilities and selecting the most probable emotion based on the probabilities. Optionally, the probabilities of the database may include performance confusion statistics. 

patent: US 7,940,914 B2

Digital FIR Filter Design using Frequency Sampling Method

In this experiment digital filter is designed using Frequency sampling Method(FSM).
If the filter is a linear phase FIR filter then Φ= -(N-1)ω/2 where N is the length of h[n].
                       The given data should be :  type of filter(LPF or HPF), Order of the filter and the digital cutoff frequency  (ω_c).If we compare it with other methods then the difference is that we find H[k] by sampling using DTFT and then we obtain transfer function h[n] by IDFT method.
In case of non linear phase filter value of Φ=0.Its simpler to implement if compared to other methods.

Digital FIR Filter Design using Windowing Method

In this experiment we used windowing method to design the digital FIR filter.
FIR filters if compared to IIR filters,its phase response is linear which is not the case in IIR filters.
In this method , FIR filter is designed by truncating infinite samples oh hd[n] using a window function. Linear phase FIR filter with impulse response h[n] is then given by h[n]=hd[n].w[n] where w[n] is an appropriate window function.
       The written program finds an appropriate window function by comparing the entered value of Stopband attenuation(As) with the stored values in the program. Then using that particular window function it calculates value of h[n].

Digital Chebyshev filter design

In this experiment, using analog filter parameters like  Passband and Stopband Attenuation, Analog Passband and Stopband Frequency as well as sampling frequency we designed digital Chebyshev Low pass and High Pass filter.
    We conclude that in Chebyshev filters ,the magnitude spectrum shows ripple in Passband and it is monotonic (no ripple) in Stopband. The advantage is for same analog filter parameters order of Chebyshev filter is less than that of Butterworth filter and thus its realization requires less hardware.

Digital Butterworth filter design

In this experiment, using analog filter parameters like  Passband and Stopband Attenuation, Analog Passband and Stopband Frequency as well as sampling frequency we designed digital Butterworth Low pass and High Pass filter.
    We conclude that the frequency bands of analog and digital filters are complex and they don't match. The advantage of this filter is that the Magnitude spectrum is monotonic in both passband as well as stop band.

OSM & OAM

OSM & OAM methods are efficient ways to evaluate discrete convolution of a very long signal & a finite impulse response of a digital filter. In OAM, Linear convolution of individual L point sequences is calculated whereas in OSM, Circular Convolution is calculated to obtain the output.In OSM, output is directly obtained by discarding few coefficients from individual stages rather than adding coefficients from OAM method.

FFT (Fast Fourier Transform)

FFT is similar to the DFT Algorithm since it is also the frequency domain representation of a signal but FFT is a fast algorithm and hence produces fast results. Cooley-Tukey Algorithm for FFT breaks a DFT signal into many similar DFTs .This reduces the computations that is why it is a fast algorithm.
The simple computations like trivial operations like multiplication by 1 are avoided producing fast results.

DFT (Discrete Fourier Transform)

DFT is basically a frequency sampling of Discrete Time Fourier Transform that produces periodic results. This is because a finite sequence of equally spaced samples of a function can be converted into equivalent length sequences of equally spaced samples. An approximate frequency spectrum of a signal can be obtained using DFT Algorithm.If the length of signal is increased then its DFT produces more compressed frequency spectrum. So we use zero padding to reduce the frequency spacing in the output spectrum. This reduces the approximation error.

Discrete Convolution and Correlation

In this experiment , Convolution and Correlation algorithms were emulated on the DSP Processor TMS320F28335.

For a Linear Time Invariant system,Convolution is basically smearing the input signal with another signal to get the output. Convolution algorithms are used in image processing for smoothing of an image. Correlation is used to measure the extent of similarity between  two signals. The value of correlation coefficient indicates the degree of similarity.