Differences
This shows you the differences between two versions of the page.
| Both sides previous revisionPrevious revisionNext revision | Previous revision | ||
| learn:courses:unit-7:start [2017/03/29 19:29] – Martha | learn:courses:unit-7:start [2022/10/04 11:38] (current) – external edit 127.0.0.1 | ||
|---|---|---|---|
| Line 1: | Line 1: | ||
| ====== Unit 7: Audio Signal Processing ====== | ====== Unit 7: Audio Signal Processing ====== | ||
| + | [[{}/ | ||
| + | |||
| === Unit 7 Labs === | === Unit 7 Labs === | ||
| - | * [[https:// | + | * [[/ |
| - | * [[https:// | + | * [[/ |
| === Download This Document === | === Download This Document === | ||
| - | * {{ : | + | {{ : |
| Line 16: | Line 18: | ||
| ===== 2. Objectives ===== | ===== 2. Objectives ===== | ||
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| Line 26: | Line 28: | ||
| ===== 3. Basic Knowledge ===== | ===== 3. Basic Knowledge ===== | ||
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| Line 37: | Line 39: | ||
| ===== 4. Equipment List ===== | ===== 4. Equipment List ===== | ||
| ==== 4.1. Hardware ==== | ==== 4.1. Hardware ==== | ||
| - | | + | |
| - | | + | |
| - | | + | |
| In addition, we suggest the following instruments: | In addition, we suggest the following instruments: | ||
| - | | + | |
| ==== 4.2. Software ==== | ==== 4.2. Software ==== | ||
| The following programs must be installed on your development workstation: | The following programs must be installed on your development workstation: | ||
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| ---- | ---- | ||
| ===== 5. Project Takeaways ===== | ===== 5. Project Takeaways ===== | ||
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| - | | + | |
| Line 102: | Line 104: | ||
| The result of the substitution of Eq. 6.2 into Eq. 6.1 is also a rational polynomial, as shown in Eq. 6.3, that relates the output, $Y(z)$, to the input, $X(z)$. The polynomial coefficients $d_j$ and $c_i$ are not the same values as $b_j$ and $a_i$ used in Eq. 6.1. | The result of the substitution of Eq. 6.2 into Eq. 6.1 is also a rational polynomial, as shown in Eq. 6.3, that relates the output, $Y(z)$, to the input, $X(z)$. The polynomial coefficients $d_j$ and $c_i$ are not the same values as $b_j$ and $a_i$ used in Eq. 6.1. | ||
| - | $$H(z) = \frac{y(z)}{X(z)} = \frac{FILTERGAIN \cdot \sum_{j-0}^{M-1} d_j \cdot z^{-1}}{1 + \sum_{i=1}^{N-1} c_j \cdot z^{-i}} = \frac{FILTERGAIN \cdot \sum_{j=0}^{M-1} d_j \cdot X(z) \cdot z^{-j}} \qquad (\text{Eq. 6.3})$$ | + | $$H(z) = \frac{y(z)}{X(z)} = \frac{FILTERGAIN \cdot \sum_{j-0}^{M-1} d_j \cdot z^{-1}}{1 + \sum_{i=1}^{N-1} c_j \cdot z^{-i}} = \frac{FILTERGAIN \cdot \sum_{j=0}^{M-1} d_j \cdot X(z) \cdot z^{-j}}{1 + \sum_{i=1}^{N-1} c_i \cdot Y(z) \cdot z^{-i}} \qquad (\text{Eq. 6.3})$$ |
| Solving Eq. 6.3 for the output, $Y(z)$, as a function of the input, $X(z)$, results in Eq. 6.4. | Solving Eq. 6.3 for the output, $Y(z)$, as a function of the input, $X(z)$, results in Eq. 6.4. | ||
| Line 110: | Line 112: | ||
| Using the relationship that the inverse $z$ operator is a delay of one sample, $X(z) \cdot z^{-i}$ becomes $x(n-i)$. Hence, Eq. 6.5 is the inverse $z$ transform of Eq. 6.3. | Using the relationship that the inverse $z$ operator is a delay of one sample, $X(z) \cdot z^{-i}$ becomes $x(n-i)$. Hence, Eq. 6.5 is the inverse $z$ transform of Eq. 6.3. | ||
| - | $$y(n) + \sum_{i=1}^{N-1 | + | $$y(n) + \sum_{i=1}^{N-1} c_i \cdot y(n-i) = FILTERGAIN \cdot \sum_{j=0}^{M-1} d_j \cdot x(n-j) \qquad (\text{Eq. 6.5})$$ |
| - | + | ||
| + | Here the nth term is the present input or most recent sample. The $(n-i)^{th}$ and $(n-j)^{th}$ terms are past outputs and inputs, respectively. Eq. 6.6 represents an equation that can be solved on a computer. This form of a digital filter is called an infinite impulse response filter because it involves past outputs in the computation. Listing A.3 in Appendix A is a C function that can implement a fourth order IIR filter. | ||
| + | |||
| + | $$y(n) = FILTERGAIN \cdot [b_0 \cdot x(n) + b_1 \cdot x(n-1) + b_2 \cdot x(n-2) + b_3 \cdot x(n-2)] - [a_1 \cdot y(n-1) + a_2 \cdot y(n-2) + a_3 \cdot y(n-3) + a_4 \cdot y(n-4)] \qquad (\text{Eq. 6.6})$$ | ||
| + | |||
| + | As will be shown below, the implementation of a digital filter is not complicated. The real science of designing digital filters lies in the selection of the design parameters that eventually produce the polynomial coefficients $b_j$ and $a_i$. While Reference 6 and 7 provide the theoretical basis for digital filter design, much of the technical background can be minimized by using filter design software programs such as the one provided by Reference 8. | ||
| + | |||
| + | === 6.3.1. FIR C Program === | ||
| + | Finite impulse response (FIR) filters are a class of digital filters that only use present and past inputs. FIR filters can be expressed by Eq. 6.5 where all ci coefficients are zero, thus resulting in Eq. 6.7. FIR filters have a finite response to any input. | ||
| + | |||
| + | $$y(n) = FILTERGAIN \cdot \sum_{j=0}^{M-1} d_j \cdot x(n-j) \qquad (\text{Eq. 6.7})$$ | ||
| + | |||
| + | Since FIR filters do not use any feedback, any rounding errors are not compounded by summed iterations, resulting in the same relative error in each consecutive calculation. FIR filters are inherently [[https:// | ||
| + | |||
| + | The main disadvantage of FIR filters is that considerably more computation power in a general purpose processor is required compared to an IIR filter with similar sharpness or [[https:// | ||
| + | |||
| + | Figure 6.2 is a screenshot of a FIR design for a low-pass filter with 2.0 kHz and 3db cutoff frequency. This was done using Iowa Hills Filter Design Software. A few comments are in order to explain the process of entering the filter specifications into the Iowa Hills filter designer. The entry box labeled OmegaC is the place where the filter 3db cutoff frequency is set. The filter specification for this example of a 16 tap FIR filter has a cutoff frequency of 2 kHz and a sampling frequency of 16 kHz. The normalized cutoff radian frequency, ωc, must first be converted to a ratio of the cutoff frequency, Fc, to the sampling frequency, Fs, as shown in Eq. 6.8. Eq. 6.9 pre-warps the cutoff frequency to compensate for frequency distortion generated by the bilinear transformation. Hence, the entry for OmegaC shown in Fig. 6.2 uses a cutoff frequency of 2111 Hz for the analog filter design algorithm so that the resulting digital filter will have a cutoff frequency of 2000 Hz.((Leland B. Jackson, Digital Filtering and Signal Processing, 3rd Ed. Kluwer Academic Publishers, ISBN 0-7923-9559-X, | ||
| + | |||
| + | $$\omega c = 2\pi FcFs \qquad (\text{Eq. 6.8})$$ | ||
| + | |||
| + | $$OmegaC = 2\pi \cdot \tan (\frac{\omega c}{2}) \qquad (\text{Eq. 6.9})$$ | ||
| + | |||
| + | {{ : | ||
| + | //Figure 6.2. 16 Tap FIR Filter response and design coefficients.// | ||
| + | |||
| + | Listing A.2 is the C code for the FIR digital filter using Q1.15 fixed point math. The Q1.15 fixed math scaled filter coefficients are generated by multiplying the coefficients listed on the right-hand window of Fig. 6.2 by 215. The code shown for fir_C_filter initially saves past inputs before implementing the algorithm using Eq. 6.7. Listing A.1 shows the code to implement the FIR filter using the MIPS DSP library function. The function setup_fir_filter is called during initialization while the function fir_MIPS_filter is called each time a new input data sample is processed. The fir_MIPS_filter function executes the FIR filter almost 7 times faster than the fir_C_filter function. | ||
| + | |||
| + | === 6.3.2 IIR C Program === | ||
| + | Infinite impulse response (IIR) filters, or recursive filters, are a class of digital filters that only use present and past inputs and past outputs. Mathematically, | ||
| + | |||
| + | The main advantage digital IIR filters have over FIR filters is their efficiency of design specifications in terms of filter prototype, bandwidth, cutoff frequency, ripple, and/or roll-off. Such a set of specifications can be met with a lower order IIR filter than would be required for an FIR filter meeting the same requirements. When implemented in a signal processor, this implies a correspondingly fewer number of calculations per time step. The computational savings is often a rather large factor. | ||
| + | |||
| + | IIR filters with [[https:// | ||
| + | |||
| + | Figure 6.3 is a screenshot of an IIR design for a low pass filter with 2.0 kHz and 3db cutoff. The same consideration for the parameter $OmegaC$ discussed in section 6.3.1 above apply here as well. The coefficients on the right are for different implementation algorithms. The $N^{th}$ order coefficients are used for implementing the algorithm shown in Eq. 6.5 and Eq. 6.6. Note that the magnitudes of the denominator coefficients are greater than unity, hence there will be an overflow if we attempt to use the Q1.15 format. However, if both the numerator and denominator coefficients are divided by two after the Q1.15 scaling, all coefficients will be less than unity and the filter will retain the desired characteristics. Halving the filter coefficients after Q1.15 scaling is the same as scaling by Q2.14. The C code for the IIR filter using these coefficients is shown in Listing A.4. | ||
| + | |||
| + | {{ : | ||
| + | //Figure 6.3. 4th order IIR Butterworth filter response and design coefficients.// | ||
| + | |||
| + | The second set of coefficients shown in Fig. 6.3 is used for a [[https:// | ||
| + | |||
| + | {{ : | ||
| + | //Figure 6.4. Flow diagram of a single stage bi-quadratic IIR digital filter.// | ||
| + | |||
| + | Figure 6.5 is a diagram of a single stage biquadratic IIR filter used by the MIPS DSP library. This unconventional model shows that there is no implementation of the $B_0$ or $A_0$ terms. Section 2.1.1.1 of the application note cited by Reference 10 explains how to convert the biquadratic coefficients shown in Fig. 6.4 to the coefficients shown in Fig. 6.5. The conversion is outlined by Eq. 6.10 through 6.12. | ||
| + | |||
| + | $$A_{i,j} = -a_{i,j} \cdot 2^{13}, i \text{and} j = 1,2 \qquad (\text{Eq. 6.10})$$ | ||
| + | |||
| + | $$B_i = \bigg( \frac{b_{i, | ||
| + | |||
| + | $$scale = \sqrt{max \big( log_2(b1_0 \cdot b2_0) \big)} \qquad (\text{Eq. 6.12})$$ | ||
| + | |||
| + | Since the scale factor is always less than or equal to 2 raised the product of b10 and b20, the gain of the filter will be less than or equal to unity. The code shown in Listing A.3 shows that the compensating gain for the example IIR filter is 2.51. Performance tests show that the IIR filter implemented using the MIP DSP library function is almost 15 times faster than using only C statements. | ||
| + | |||
| + | {{ : | ||
| + | //Figure 6.5. Flow diagram of a single stage MIPS IIR Biquadratic filter.// | ||
| + | |||
| + | Figure 6.6 plots the response of the four filter algorithms for Listing A.1 through A.4. Only three plots are distinguishable because the results of the input X and the IIR1 corresponding to Listing A.1, as well as the FIR filter result for FIR1 are nearly identical. | ||
| + | |||
| + | {{ : | ||
| + | //Figure 6.6. Plot of outputs of the four example filters to a 1 kHz input sine wave.// | ||
| + | |||
| ---- | ---- | ||
| + | |||
| + | ===== 7. Lab Exercises ===== | ||
| + | As modeled in Fig. 7.1, the blocks shaded blue represent analog circuits and components. The blocks shaded green represent the hardware resources within the microprocessor itself. The electromechanical transducers are the speaker in Fig. 7.1 and the microphone in Fig. 7.2. | ||
| + | |||
| + | Figure 7.1 is the block diagram for Lab 7a that implements a sine wave generator using an IIR filter. The frequency of the synthesized sine wave is selected by setting one of the eight slide switches on the Basys MX3 board high. The frequencies synthesized range from 500 to 7500 Hz in steps of 1000 Hz. Refer to Lab 7a for details concerning the algorithms to synthesize a sine wave and project specifications. The Basys MX3 board LCD will display the selected frequency and the on-board amplifier-speaker circuit shown in Fig. B.1 of Appendix B. | ||
| + | |||
| + | {{ : | ||
| + | //Figure 7.1. Block diagram of a digital signal generator (Lab 7a).// | ||
| + | |||
| + | Figure 7.2 is the block diagram for Lab 7b that implements a frequency spectrum analyzer using discrete Fourier transforms (DFT) implemented by an FIR type algorithm. One DFT algorithm will be programmed using conventional C statements while a second DFT algorithm will use a MIPS DSP library function. The Basys MX3 board LCD will input the audio signal using the Basys MX3 microphone circuit, shown in Fig. B.2, and will display the frequency spectrum on the on-board LCD. | ||
| + | |||
| + | {{ : | ||
| + | //Figure 7.2. Block diagram of a digital spectrum analyzer (Lab 7b).// | ||
| + | |||
| + | |||
| + | ---- | ||
| + | |||
| + | ===== 8. References ===== | ||
| + | - Basys MX3 Trainer Board Reference Manual | ||
| + | - [[http:// | ||
| + | - PIC32 Family Reference Manual, Timers Section 14: [[http:// | ||
| + | - [[http:// | ||
| + | - [[http:// | ||
| + | - Steven W. Smith, __The Scientist and Engineer’s Guide to Digital Signal Processing__, | ||
| + | - “Design of Digital Filters”, [[https:// | ||
| + | - Iowa Hills Software Digital and Analog Filters, [[http:// | ||
| + | - “A Discrete Fourier Transform Based Digital DTMF Detection Algorithm”, | ||
| + | - Application Report SPRA867, “Parametric Equalization on TMS320C6000 DSP”, [[http:// | ||
| + | |||
| + | |||
| + | ---- | ||
| + | |||
| + | ===== Appendix A: Filter C Code ===== | ||
| + | ==== Listing A.1. C Code to Implement a 16 tap FIR Digital Filter Using MIPS FIR Function ==== | ||
| + | < | ||
| + | #define K 16 | ||
| + | int16_t delayline[K] = {0}; | ||
| + | int16_t coeffs2x[2*K] = {0}; | ||
| + | |||
| + | void setup_fir_filter(void) | ||
| + | { | ||
| + | int i; | ||
| + | for (i = 0; i < K; i++) | ||
| + | delayline[i] = 0; | ||
| + | // lowpass2KHzFIRCoefs declared in Listing 3 | ||
| + | | ||
| + | |||
| + | } | ||
| + | |||
| + | int16_t fir_MIPS_filter(int16_t indata) | ||
| + | { | ||
| + | int i; | ||
| + | int15_t outdata; | ||
| + | |||
| + | | ||
| + | |||
| + | | ||
| + | } | ||
| + | </ | ||
| + | |||
| + | ==== Listing A.2. C Code to Implement a 16 tap FIR Digital Filter Using Fixed Point Math ==== | ||
| + | < | ||
| + | int16_t lowpass2KHzFIRCoefs[16]= | ||
| + | { | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | | ||
| + | }; | ||
| + | |||
| + | int fir_C_filter(int indata) | ||
| + | { | ||
| + | #define FIR_ORDER 16 // Filter order | ||
| + | /* | ||
| + | Fixed point math uses Q1.15 format to scale coefficients | ||
| + | */ | ||
| + | static int x[FIR_ORDER] = {0}; // Input memory array initialized to zero | ||
| + | static int y[FIR_ORDER] = {0}; // Output memory array initialized to zero | ||
| + | int i,j; | ||
| + | for(j = FIR_ORDER; j > 0; j--) // Time shift past inputs and outputs | ||
| + | { | ||
| + | x[j] = x[j-1]; | ||
| + | y[j] = y[j-1]; | ||
| + | } | ||
| + | x[0] = indata; | ||
| + | y[0] = (x[0] * lowpass2KHzFIRCoefs[0]); | ||
| + | for(j = 1; j < FIR_ORDER; j++) | ||
| + | | ||
| + | y[0] += (x[j] * lowpass2KHzFIRCoefs[j]); | ||
| + | } | ||
| + | y[0] /= Q15; // Normalize the result | ||
| + | | ||
| + | } | ||
| + | </ | ||
| + | |||
| + | ==== Listing A.3. C Code to Implement a Fourth Order IIR Digital Filter Using MIPS IIR Function ==== | ||
| + | < | ||
| + | #define B 2 | ||
| + | |||
| + | biquad16 bq[B]; | ||
| + | int16_t coeffs[4*B]; | ||
| + | int16_t delayline[2*B]; | ||
| + | |||
| + | void setup_iir_filter (void) | ||
| + | { | ||
| + | int i; | ||
| + | // | ||
| + | for (i = 0; i < 2*B; i++) | ||
| + | | ||
| + | |||
| + | bq [0].a1 = 6590; //a( 1,1)/2 | ||
| + | bq [0].a2 = -1556; | ||
| + | bq [0].b1 = 2048; //b( 1,1)/2 | ||
| + | bq [0].b2 = 1024; //b( 1,2)/2 | ||
| + | |||
| + | bq [1].a1 = 8640; //a( 2,1)/2 | ||
| + | bq [1].a2 = -4588; | ||
| + | bq [1].b1 = 2048; //b( 2,1)/2 | ||
| + | bq [1].b2 = 1024; //b( 2,2)/2 | ||
| + | |||
| + | | ||
| + | } | ||
| + | |||
| + | int16_t iir_filter (int16_t indata) | ||
| + | { | ||
| + | int i; | ||
| + | int16_t outdata; | ||
| + | int gain = 2.51; | ||
| + | |||
| + | //calculate iir filter biquad | ||
| + | | ||
| + | | ||
| + | } | ||
| + | </ | ||
| + | |||
| + | ==== Listing A.4. C Code to Implement a Fourth Order IIR Digital Filter Using Fixed Point Math ==== | ||
| + | < | ||
| + | int iir_filter(void) | ||
| + | { | ||
| + | #define IIR_ORDER 4 // Filter order | ||
| + | /* | ||
| + | Constants b0 – b3 and a1 – a3 as well as FILTERGAIN are defined as Q1.13 scaled signed integer 16 data types. | ||
| + | |||
| + | */ | ||
| + | #define Q14 1<< | ||
| + | static int x[IIR_ORDER] = {0}; // Past input memory array initialized to zero | ||
| + | static int y[IIR_ORDER] = {0}; // Past output memory array initialized to zero | ||
| + | int i; | ||
| + | |||
| + | // propagate past inputs and outputs | ||
| + | for(i = ORDER; i > 0; i--) | ||
| + | { | ||
| + | x[i] = x[i-1]; | ||
| + | y[i] = y[i-1]; | ||
| + | } | ||
| + | x[0] = sample; | ||
| + | y[0] = x[0] * b0; // Compute the first numerator term | ||
| + | |||
| + | // Compute the remaining filter terms | ||
| + | for(i = 1; i < ORDER; i++) | ||
| + | { | ||
| + | |||
| + | y[0] += ((x[i] * b[i]) * FILTER_GAIN) - (y[i] * a[i]); | ||
| + | } | ||
| + | y[0] /= Q14; // Normalize the result | ||
| + | | ||
| + | } | ||
| + | </ | ||
| + | |||
| + | |||
| + | ---- | ||
| + | |||
| + | |||
| + | ===== Appendix B: Audio Amplifiers ===== | ||
| + | {{ : | ||
| + | //Figure B.1. Basys MX3 Trainer Board Audio Output Schematic Diagram.// | ||
| + | |||
| + | {{ : | ||
| + | //Figure B.2. Basys MX3 Microphone Schematic.// | ||
| + | |||
| + | ---- | ||
| + | |||
| + | [[{}/ | ||
| + | [[{}/ | ||
| + | [[{}/ | ||
| + | [[{}/ | ||
| {{tag> | {{tag> | ||