dsp.HDLCICDecimation

Decimate signal using cascaded integrator-comb filter — optimized for HDL code generation

Description

The dsp.HDLCICDecimation System object™ decimates an input signal by using a cascaded integrator-comb (CIC) decimation filter. CIC filters are a class of linear phase FIR filters consisting of a comb part and an integrator part. The CIC decimation filter structure consists of N sections of cascaded integrators, a rate change factor of R, and then N sections of cascaded comb filters. For more information about CIC decimation filters, see Algorithms.

The System object supports fixed and variable decimation rates for scalar inputs and only fixed decimation for vector inputs. For both types of inputs, the System object provides a scalar output. The System object provides an architecture suitable for HDL code generation and hardware deployment.

The System object supports real and complex fixed-point inputs.

To filter input data with an HDL-optimized CIC decimation filter:

1. Create the dsp.HDLCICDecimation object and set its properties.

2. Call the object with arguments, as if it were a function.

To learn more about how System objects work, see What Are System Objects?.

Creation

Syntax

cicDecFilt = dsp.HDLCICDecimation

cicDecFilt = dsp.HDLCICDecimation(Name,Value)

Description

cicDecFilt = dsp.HDLCICDecimation creates an HDL-optimized CIC decimation filter System object, cicDecFilt, with default properties.

cicDecFilt = dsp.HDLCICDecimation(Name,Value) creates the filter with properties set using one or more name-value pairs. Enclose each property name in single quotes.

Properties

Unless otherwise indicated, properties are nontunable, which means you cannot change their values after calling the object. Objects lock when you call them, and the release function unlocks them.

If a property is tunable, you can change its value at any time.

For more information on changing property values, see System Design in MATLAB Using System Objects.

VariableDownsample — Variable decimation rate

(0) false (default) | (1) true

Set this property to true (1) to operate the System object with a variable decimation rate specified by the decimFactor input argument. Set this property to false (0) to operate the object with a fixed decimation rate, specified by the DecimationFactor property.

For vector inputs, the object does not support variable decimation rate.

DecimationFactor — Decimation factor

2 (default) | integer from 2 to 2048

Specify the decimation factor as an integer from 2 to 2048. This value represents the rate with which you want to decimate the input.

When you set the VariableDownsample property to true, this property sets the upper bound of the range of valid values for the decimFactor argument.

DifferentialDelay — Differential delay

1 (default) | 2

Specify the differential delay of the comb part of the filter as either 1 or 2 cycles.

NumSections — Number of integrator or comb sections

2 (default) | integer from 1 to 6

Specify the number of sections in either the integrator or the comb part of the filter as an integer from 1 to 6.

OutputDataType — Data type of output

'Full precision' (default) | 'Same word length as input' | 'Minimum section word lengths'

Choose the data type of the filtered output data.

• 'Full precision' — The output data type has a word length equal to the input word length plus gain bits.

• 'Same word length as input' — The output data type has a word length equal to the input word length.

• 'Minimum section word lengths' — The output data type uses the word length you specify in the OutputWordLength property. When you choose this option, the System object applies a Pruning algorithm internally. For more information about Pruning, see Output Data Type. This option is not supported when VariableDownsample is true.

OutputWordLength — Word length of output

16 (default) | integer from 2 to 104

Word length of the output, specified as an integer from 2 to 104.

Note

When this value is less than 7 the output data values might overflow.

Dependencies

To enable this property, set the OutputDataType property to 'Minimum section word lengths'.

GainCorrection — Compensate output gain

false (default) | true

Set this property to true to compensate for the output gain of the filter.

Depending on the type of decimation you specify and the value of this property, the latency of the object changes. Here, N means the number of sections and vecLen means the length of the vector.

For a scalar input with fixed decimation (VariableDownsample is false):

• With gain correction off, the latency of the object is 3 + N clock cycles.

• With gain correction on, the latency of the object is 3 + N + 9 clock cycles.

For a scalar input with variable decimation (VariableDownsample is true):

• With gain correction off, the latency of the object is 4 + N clock cycles.

• With gain correction on, the latency of the object is 4 + N + 9 clock cycles.

For a vector input with fixed decimation (VariableDownsample is false):

• With gain correction off, the latency of the object is floor((vecLen – 1) * (N/vecLen)) + 1 + N + (2 + (vecLen + 1) * N clock cycles.

• With gain correction on, the latency of the object is floor((vecLen – 1) * (N/vecLen)) + 1 + N + (2 + (vecLen + 1) * N) + 9 clock cycles.

Note

For vector inputs, the object does not support variable decimation.

ResetIn — Enable reset argument

false (default) | true

When you set this property to true, the System object expects a reset input argument.

Usage

Syntax

[dataOut,validOut] = cicDecFilt(dataIn,validIn)

[dataOut,validOut] = cicDecFilt(dataIn,validIn,decimFactor)

[dataOut,validOut] = cicDecFilt(dataIn,validIn,reset)

[dataOut,validOut] = cicDecFilt(dataIn,validIn,decimFactor,reset)

Description

[dataOut,validOut] = cicDecFilt(dataIn,validIn) filters and decimates the input data using a fixed decimation factor only when validIn is true.

[dataOut,validOut] = cicDecFilt(dataIn,validIn,decimFactor) filters the input data using the specified variable decimation factor, decimFactor. The VariableDownsample property must be set to true.

[dataOut,validOut] = cicDecFilt(dataIn,validIn,reset) filters the input data when reset is false and clears filter internal states when reset is true. The System object expects the reset argument only when you set the ResetIn property to true.

[dataOut,validOut] = cicDecFilt(dataIn,validIn,decimFactor,reset) filters the input data when reset is false and clears filter internal states when reset is true. The System object expects the reset argument only when you set the ResetIn property to true. The VariableDownsample property is set to true.

Input Arguments

dataIn — Input data

scalar | column vector

Specify input data as a scalar or a column vector of length 1 to 64. The input data must be a signed integer or signed fixed point with a word length less than or equal to 32. DecimationFactor property must be an integer multiple of the input frame size.

Data Types: int8 | int16 | int32 | fi
Complex Number Support: Yes

validIn — Indication of valid input data

logical scalar

Control signal that indicates if the input data is valid.

When validIn is 1 (true), the System object captures the value from the dataIn input argument. When validIn is 0 (false), the System object ignores the dataIn input value.

Data Types: logical

decimFactor — Variable decimation rate

scalar

Specifies the decimation rate.

The decimFactor value must be of data type ufix12 data type and an integer in the range from 2 to the DecimationFactor property value.

Dependencies

To enable this argument, set the VariableDownsample property to true.

Data Types: fi(0,12,0)

reset — Clear internal states

logical scalar

Clear internal states, specified as a logical scalar.

When this value is 1 (true), the System object stops the current calculation and clears all internal states. When this value is 0 (false) and validIn is 1 (true), the System object starts a new filtering operation.

Dependencies

To enable this argument, set the ResetIn property to true.

Data Types: logical

Output Arguments

dataOut — CIC-decimated output data

scalar

CIC decimated output data, returned as a scalar.

The OutputDataType property sets the output data type of this argument. See OutputDataType.

Data Types: int8 | int16 | int32 | fi
Complex Number Support: Yes

validOut — Indication of valid output data

logical scalar

Control signal that indicates if the data from the dataOut output argument is valid. When this value is 1 (true), the System object returns valid data from the dataOut output argument. When this value is 0 (false), the values of the dataOut output argument are not valid.

Data Types: logical

Object Functions

To use an object function, specify the System object as the first input argument. For example, to release system resources of a System object named obj, use this syntax:

release(obj)

Specific to dsp.HDLCICDecimation

getLatency

Latency of CIC decimation filter

Common to All System Objects

step	Run System object algorithm
release	Release resources and allow changes to System object property values and input characteristics
reset	Reset internal states of System object

Examples

Create CIC Decimation Filter for HDL Code Generation

This example shows how to use a dsp.HDLCICDecimation System object™ to filter and downsample the data. This object supports scalar and vector inputs. In this example, two functions are provided to work with scalar and vector input data separately. You can generate the HDL code from these functions.

Generate Frames of Random Input Samples

Set up these workspace variables for the object to use. The object supports fixed and variable decimation rates for scalar inputs and only a fixed decimation rate for vector inputs. The example runs the HDLCIC_maxR8 function when you set the scalar variable to true and runs the HDLCIC_vec function when you set the scalar variable to false. For scalar inputs, choose a range of the input varRValue values and set the decimation factor value, R, to the maximum expected decimation factor. For vector inputs, the input data must be a column vector of size 1 to 64 and R must be an integer multiple of the input frame size.

R = 8; % decimation factor
M = 1; % differential delay
N = 3; % number of sections
scalar = false; % true for scalar; false for vector
if scalar
    varRValue = [4, 5, 6, 7, 8];
    vecSize = 1;
else
    varRValue = R;
    vecSize = randsrc(1,1,factor(R));
end

numFrames = length(varRValue);
dataSamples = cell(1,numFrames);
varRtemp = cell(1,numFrames);
framesize = zeros(1,numFrames);
refOutput = [];
WL = 0; % Word length
FL = 0; % Fraction length

Generate Reference Output from dsp.CICDecimation System Object™

Generate frames of random input samples and apply the samples to the dsp.CICDecimation System object. The output generated from this System object is used as a reference data for comparison. The System object does not support a variable decimation rate, so you must create and release the object for each change in decimation factor value.

totalsamples = 0;
for i = 1:numFrames
    framesize(i) = varRValue(i)*randi([5 20],1,1);
    dataSamples{i} = fi(randn(vecSize,framesize(i)),1,16,8);
    ref_cic = dsp.CICDecimator('DifferentialDelay',M,...
        'NumSections',N,...
        'DecimationFactor',varRValue(i));
    refOutput = [refOutput,ref_cic(dataSamples{i}(:)).'];
    release(ref_cic);
end

Run a Function that contains dsp.HDLCICDecimation System Object

Set the properties of the System object to match the input data parameters and run the respective function based on the input type. These functions operate on a stream of data samples rather than a frame. You can generate HDL code from these functions.

The example uses the HDLCIC_maxR8 function for a scalar input.

function [dataOut,validOut] = HDLCIC_maxR8(dataIn,validIn,R)
%HDLCIC_maxR8
% Performs CIC decimation with an input decimation factor up to 8. 
% sampleIn is a scalar fixed-point value.
% validIn is a logical scalar value.
% You can generate HDL code from this function.

  persistent cic8;
  if isempty(cic8)
    cic8 = dsp.HDLCICDecimation('DecimationFactor',8,...
                                'VariableDownsample',true,...
                                'DifferentialDelay',1,...
                                'NumSections',3);
  end    
  [dataOut,validOut] = cic8(dataIn,validIn,R);
end

The example uses the HDLCIC_vec function for a vector input.

function [dataOut,validOut] = HDLCIC_vec(dataIn,validIn)
%HDLCIC_vec
% Performs CIC decimation with an input vector. 
% sampleIn is a fixed-point vector.
% validIn is a logical scalar value.
% You can generate HDL code from this function.

  persistent cicVec;
  if isempty(cicVec)
    cicVec = dsp.HDLCICDecimation('DecimationFactor',8,...
                                'VariableDownsample',false,...
                                'DifferentialDelay',1,...
                                'NumSections',3);
  end    
  [dataOut,validOut] = cicVec(dataIn,validIn);
end

To flush remaining data, run the object by inserting the required number of idle cycles after each frame through latency variable. For more information, see GainCorrection property.

Initialize the output to a size large enough to accommodate the output data. The final size is expected to be smaller than totalsamples due to decimation.

latency = floor((vecSize - 1)*(N/vecSize))+ 1+ N +(2+(vecSize + 1)*N)+ 9;
dataOut = zeros(1,totalsamples+numFrames*latency);
validOut = zeros(1,totalsamples+numFrames*latency);
idx=0;
for ij = 1:numFrames
    if scalar
        % scalar input with variable decimation
        for ii = 1:length(dataSamples{ij})
            idx = idx+1;
            [dataOut(idx),validOut(idx)] = HDLCIC_maxR8(...
                dataSamples{ij}(ii),...
                true,...
                fi(varRValue(ij),0,12,0));
        end
        for ii = 1:latency
            idx = idx+1;
            [dataOut(idx),validOut(idx)] = HDLCIC_maxR8(...
                fi(0,1,16,8),...
                false,...
                fi(varRValue(ij),0,12,0));
        end

    else
        % vector input with fixed decimation
        for ii = 1:size(dataSamples{ij},2)
            idx = idx+1;
            [dataOut(idx),validOut(idx)] = HDLCIC_vec(...
                dataSamples{ij}(:,ii),...
                true);
        end
        for ii = 1:latency
            idx = idx+1;
            [dataOut(idx),validOut(idx)] = HDLCIC_vec(...
                fi(zeros(vecSize,1),1,16,8),...
                false);
        end
    end
end

Compare the Function Output with the Reference Data

Compare the function results against the output from the dsp.CICDecimation object.

cicOutput = dataOut(validOut==1);

fprintf('\nHDL CIC Decimation\n');
difference = (abs(cicOutput-refOutput(1:length(cicOutput)))>0);
fprintf('\nTotal number of samples differed between Behavioral and HDL simulation: %d \n',sum(difference));

HDL CIC Decimation

Total number of samples differed between Behavioral and HDL simulation: 0 

Explore Latency of HDL CIC Decimation Object

The latency of the dsp.HDLCICDecimation System object™ varies depending on how many integrator and comb sections your filter has, input vector size, and whether you enable gain correction. Use the getLatency function to find the latency of a particular filter configuration. The latency is the number of cycles between the first valid input and the first valid output, assuming the input is continuously valid.

Create a dsp.HDLCICDecimation System object™ and request the latency. The default filter has two sections, and gain correction is disabled.

hdlcic = dsp.HDLCICDecimation

hdlcic = 
  dsp.HDLCICDecimation with properties:

    VariableDownsample: false
      DecimationFactor: 2
     DifferentialDelay: 1
           NumSections: 2
        OutputDataType: 'Full precision'
        GainCorrection: false
               ResetIn: false

L_def = getLatency(hdlcic)

L_def = 5

Modify the filter object to have three integrator and comb sections. Check the resulting change in latency.

hdlcic.NumSections = 3;
L_3sec = getLatency(hdlcic)

L_3sec = 6

Enable the gain correction on the filter object with vector input size 2. Check the resulting change in latency.

hdlcic.GainCorrection = true;
vecSize = 2;
L_wgain = getLatency(hdlcic,vecSize)

L_wgain = 25

Algorithms

CIC Decimation Filter

The transfer function of a CIC decimation filter is

$$H(z)={\left[{\displaystyle \sum _{k=0}^{RM-1}{z}^{-k}}\right]}^N=\frac{{\left(1-z^{-RM}\right)}^N}{{\left(1-z^{-1}\right)}^N}=\frac{1}{{\left(1-z^{-1}\right)}^N}·\frac{{\left(1-z^{-RM}\right)}^N}{1}=H_{\text{I}}{}^N(z)·H_{\text{c}}{}^N(z),$$

where

• H_I is the transfer function of the integrator part of the CIC filter.

• H_C is the transfer function of the comb part of the CIC filter.

• N is the number of sections. The number of sections in a CIC filter is defined as the number of sections in either the comb part or the integrator part of the filter. This value does not represent the total number of sections throughout the entire filter.

• R is the decimation factor.

• M is the differential delay.

CIC Filter Structure

The dsp.HDLCICDecimation System object has the CIC filter structure shown in this figure. The structure consists of N sections of cascaded integrators, a rate change factor of R, and N sections of cascaded comb filters [1].

Designs can put the unit delay in the integrator part of the CIC filter in either the feed-forward or feedback path. These two configurations yield an identical filter frequency response. However, the numerical outputs from these two configurations are different due to the latency of the paths. This System object puts the unit delay in the feed-forward path of the integrator.

Fixed and Variable Decimation

The System object downsamples the integrator stage output using R, either the fixed decimation rate provided using the DecimationFactor property or the variable decimation rate provided using the decimFactor input argument. At the downsampler stage, the System object uses a counter to count the valid input samples, which depend on the decimation rate. Whenever the decimation rate changes, the object resets and starts a new calculation from the next sample. This mechanism prevents the System object from accumulating invalid values. Then, the System object provides the decimated output to the comb part.

Gain Correction

The gain of the System object is given by $$Gain={(R\text{x}M)}^N$$.

where:

• R is the DecimationFactor property value.

• M is the DifferentialDelay property value.

• N is the NumSections property value.

The System object implements gain correction in two parts: coarse gain and fine gain. In coarse gain correction, the System object calculates the shift value, adds the shift value to the fractional bits to create a numeric type, and then performs bit-shift left. In fine gain correction, the System object divides the remaining gain with the coarse gain if the gain is not a power of 2 and then multiplies the coarse gain corrected value with the inverse value of fine gain. All possible shift and fine gain values are precalculated initially and stored in an array before the System object starts processing.

You can modify this equation as $$Gain=2^{cGain}\text{x}fGain$$, where cGain means coarse gain and fGain means fine gain.

• $$cGain=floor({\log }_{2}Gain)$$

• $$fGain=Gain/2^{cGain}=Gain\text{x}{2}^{-cGain}{}^{}$$

To perform GainCorrection when the VariableDownsample property is enabled, the System object sets the output data type configured with the maximum decimation rate and bit-shifts left for all the values under the maximum decimation rate. The bit-shift value is equal to $${\text{Maximum gain - log}}_{\text{2}}\text{(current gain)}$$.

Output Data Type

This section explains how the System object determines the output data type. For example, consider a filter with DecimationFactor, DifferentialDelay, and NumSections values 8, 1, and 3, respectively, with an input width of 16 bits.

The output word length is calculated as: $$B_{\text{OUT}}=B_{\text{IN}}+[{\log }_2(Gain)]$$,

where:

• $$Gain={(R\text{x}M)}^N$$

• B_IN is the input word length.

• B_OUT is the output word length.

When you set the OutputDataType property to 'Full precision', the System object returns data with a word length of 25 bits, by adding nine gain bits to the input word length.

When you set the OutputDataType property to 'Same word length as input', the object outputs data with a word length of 16, which is the same length as the input word length. The internal integrator and comb stages use the full-precision data type with 25 bits.

When you set the OutputDataType property to 'Minimum section word lengths' and the OutputWordLength property to 16, the System object returns data with a word length of 16 bits. In this case, the object changes the bit width at each stage, based on the Pruning algorithm.

If the OutputWordLength property value is less than the number of bits requested at the output, the least significant bits (LSBs) at the earlier stages are pruned. The Hogenauer algorithm provides the number of LSBs to discard at each stage. This algorithm minimizes the loss of information in the output data [1].

Latency

This section shows the latencies of the System object for a scalar input when the System object is operated with fixed and variable decimation rates and for a vector input when the System object is operated with a fixed decimation rate.

Scalar Input

This figure shows the output of the System object for the default configuration, that is, with a fixed decimation rate and DecimationFactor, DifferentialDelay, and NumSections values of 2, 1, and 2, respectively. The System object returns valid output data at every second cycle based on the fixed DecimationFactor value of 2. The latency of the System object is 5 clock cycles, calculated as 3 + N.

This figure shows the output of the System object with a fixed decimation rate, DecimationFactor, DifferentialDelay, and NumSections values of 8, 1, and 3, respectively and GainCorrection set to true. The System object returns valid output data at every eighth cycle based on the fixed DecimationFactor value of 8. The latency of the object is 15 clock cycles and is calculated as 3 + N + 9.

This figure shows the output of the System object for variable decimFactor values 2, 4, and 8 along with M and N values 1 and 3. The GainCorrection property is set to false. The System object returns valid output data at the second, fourth, and eighth cycles corresponding to the decimFactor values 2, 4, and 8, respectively. The System object accepts decimFactor argument value changes only when input validIn is 1 (true). The latency of the System object is 7 clock cycles, calculated as 4 + N.

Vector Input

The latency of the System object for a vector input is calculated using this formula: floor((vecLen – 1) * (N/vecLen)) + 1 + N + 9 * GainCorrection + (2 + (vecLen + 1) * N), where vecLen is the length of the vector and N is the number of sections.

This figure shows the output of the System object for a two-element column vector input with default configuration, that is, with a fixed decimation rate and DecimationFactor, DifferentialDelay, and NumSections values of 2, 1, and 2, respectively. The latency of the block is 12 clock cycles.

This figure shows the output of the System object for an eight-element column vector input with a fixed decimation rate and R, M, and N values of 8, 1, and 3, respectively and GainCorrection set to true. The latency of the block is 44 clock cycles.

Performance

The performance of the synthesized HDL code varies with your target and synthesis options. This table shows the resource and performance data synthesis results of the block for a scalar input with fixed and variable decimation rates and for a two-element column vector input with fixed decimation rate DecimationFactor, DifferentialDelay, and NumSections values are 2, 1, and 2, respectively. The generated HDL is targeted to the Xilinx^® Zynq^®-7000 ZC706 evaluation board.

Input Data	Decimation Type	Slice LUTs	Slice Registers	Maximum Frequency in MHz
Scalar	Fixed rate	101	166	711.74
	Variable rate	206	186	441.70
Vector	Fixed rate	218	627	624.61

The resources and frequencies vary based on the type of input data and R, M, and N values and other selected property values. Using a vector input can increase the throughput, however this option also increases the number of hardware resources that the System object uses.

References

[1] Hogenauer, E. “An Economical Class of Digital Filters for Decimation and Interpolation.” IEEE Transactions on Acoustics, Speech, and Signal Processing 29, no. 2 (April 1981): 155–62. https://doi.org/10.1109/TASSP.1981.1163535.