dsp.Channelizer

Polyphase FFT analysis filter bank

Description

The dsp.Channelizer System object™ separates a broadband input signal into multiple narrow subbands using a fast Fourier transform (FFT)-based analysis filter bank. The filter bank uses a prototype lowpass filter and is implemented using a polyphase structure. You can specify the filter coefficients directly or through design parameters.

To separate a broadband signal into multiple narrow subbands:

1. Create the dsp.Channelizer 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? (MATLAB).

Creation

Syntax

channelizer = dsp.Channelizer

channelizer = dsp.Channelizer(M)

channelizer = dsp.Channelizer(Name,Value)

Description

channelizer = dsp.Channelizer creates a polyphase FFT analysis filter bank System object that separates a broadband input signal into multiple narrowband output signals. This object implements the inverse operation of the dsp.ChannelSynthesizer System object.

channelizer = dsp.Channelizer(M) creates an M-band polyphase FFT analysis filter bank, with the NumFrequencyBands property set to M.

Example: channelizer = dsp.Channelizer(16);

channelizer = dsp.Channelizer(Name,Value) creates a polyphase FFT analysis filter bank with each specified property set to the specified value. Enclose each property name in single quotes.

Example: channelizer = dsp.Channelizer('NumTapsPerBand',20,'StopbandAttenuation',140);

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 (MATLAB).

NumFrequencyBands — Number of frequency bands

8 (default) | positive integer greater than 1

Number of frequency bands into which the object separates the input broadband signal, specified as a positive integer greater than 1. This property corresponds to the number of polyphase branches and the FFT length used in the filter bank.

Example: 16

Example: 64

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

OversamplingRatio — Oversampling ratio

1 (default) | positive integer

Oversampling ratio, specified as a positive scalar divisor of the number of frequency bands.

If the value is greater than 1, the output sample rate is different from the channel spacing. The channelizer is then known as the non-maximally decimated channelizer or oversampled channelizer. For more details, see Algorithm.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Specification — Filter design parameters or coefficients

'Number of taps per band and stopband attenuation' (default) | 'Coefficients'

Filter design parameters or filter coefficients, specified as one of these options:

• 'Number of taps per band and stopband attenuation' — Specify the filter design parameters through the NumTapsPerBand and Stopband attenuation (dB) properties.

• 'Coefficients' — Specify the filter coefficients directly using the LowpassCoefficients property.

NumTapsPerBand — Number of filter coefficients per frequency band

12 (default) | positive integer

Number of filter coefficients each polyphase branch uses, specified as a positive integer. The number of polyphase branches matches the number of frequency bands. The total number of filter coefficients for the prototype lowpass filter is given by NumFrequencyBands × NumTapsPerBand. For a given stopband attenuation, increasing the number of taps per band narrows the transition width of the filter. As a result, there is more usable bandwidth for each frequency band at the expense of increased computation.

Example: 8

Example: 16

Dependencies

This property applies when you set Specification to 'Number of taps per band and stopband attenuation'.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

StopbandAttenuation — Stopband attenuation

80 (default) | positive real scalar

Stopband attenuation of the lowpass filter, specified as a positive real scalar in dB. This value controls the maximum amount of aliasing from one frequency band to the next. When the stopband attenuation increases, the passband ripple decreases. For a given stopband attenuation, increasing the number of taps per band narrows the transition width of the filter. As a result, there is more usable bandwidth for each frequency band at the expense of increased computation.

Example: 80

Dependencies

This property applies when you set Specification to 'Number of taps per band and stopband attenuation'.

Data Types: single | double

LowpassCoefficients — Coefficients of prototype lowpass filter

[1×49 double] (default) | row vector

Coefficients of the prototype lowpass filter, specified as a row vector. The default vector of coefficients is obtained using rcosdesign(0.25,6,8,'sqrt'). There must be at least one coefficient per frequency band. If the length of the lowpass filter is less than the number of frequency bands, the object zero-pads the coefficients.

If you specify complex coefficients, the object designs a prototype filter that is centered at a nonzero frequency, also known as a bandpass filter. The modulated versions of the prototype bandpass filter appear with respect to the prototype filter and are wrapped around the frequency range [−F_s F_s]. For an example, see Channelizer with Complex Coefficients.

Tunable: Yes

Dependencies

This property applies when you set Specification to 'Coefficients'.

Data Types: single | double
Complex Number Support: Yes

Usage

Syntax

channOut = channelizer(input)

Description

channOut = channelizer(input) separates the broadband input signal into a number of narrow band signals contained in the columns of the channelizer output.

Input Arguments

input — Data input

vector | matrix

Data input, specified as a vector or a matrix. The number of rows in the input signal must be a multiple of the number of frequency bands of the filter bank. Each column of the input corresponds to a separate channel. If M is the number of frequency bands, and the input is an L-by-1 matrix, then the output signal has dimensions L/M-by-M. Each narrowband signal forms a column in the output. If the input has more than one channel, that is, it has dimensions L-by-N with N > 1, then the output has dimensions L/M-by-M-by-N.

This object supports variable-size input signals. You can change the input frame size (number of rows) even after calling the algorithm. However, the number of channels (number of columns) must remain constant.

Example: randn(64,4)

Data Types: single | double
Complex Number Support: Yes

Output Arguments

channOut — Channelizer output

matrix | 3-D array

Channelizer output, returned as a matrix or a 3-D array. If the input is an L-by-1 matrix, then the output signal has dimensions L/M-by-M, where M is the number of frequency bands. Each narrowband signal forms a column in the output. If the input has more than one channel, that is, it has dimensions L-by-N with N > 1, then the output has dimensions L/M-by-M-by-N.

Data Types: single | double
Complex Number Support: Yes

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.Channelizer

coeffs	Coefficients of prototype lowpass filter
tf	Return transfer function of overall prototype lowpass filter
polyphase	Return polyphase matrix
freqz	Frequency response of filters in channelizer
fvtool	Visualize the filters in the channelizer
bandedgeFrequencies	Compute the bandedge frequencies
centerFrequencies	Compute center frequencies
getFilters	Return matrix of channelizer FIR filters

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

Quadrature Mirror Filter Bank

The quadrature mirror filter bank (QMF) contains an analysis filter bank section and a synthesis filter bank section. dsp.Channelizer implements the analysis filter bank. dsp.ChannelSynthesizer implements the synthesis filter bank using the efficient polyphase implementation based on a prototype lowpass filter.

Initialization

Initialize the dsp.Channelizer and dsp.ChannelSynthesizer System objects. Each object is set up with 8 frequency bands, 8 polyphase branches in each filter, 12 coefficients per polyphase branch, and a stopband attenuation of 140 dB. Use a sine wave with multiple frequencies as the input signal. View the input spectrum and the output spectrum using a spectrum analyzer.

offsets = [-40,-30,-20,10,15,25,35,-15];
sinewave = dsp.SineWave('ComplexOutput',true,'Frequency',...
    offsets+(-375:125:500),'SamplesPerFrame',800);

channelizer = dsp.Channelizer('StopbandAttenuation',140);
synthesizer = dsp.ChannelSynthesizer('StopbandAttenuation',140);
spectrumAnalyzer =  dsp.SpectrumAnalyzer('ShowLegend',true,'NumInputPorts',...
    2,'ChannelNames',{'Input','Output'},'Title','Input and Output of QMF');

Streaming

Use the channelizer to split the broadband input signal into multiple narrow bands. Then pass the multiple narrowband signals into the synthesizer, which merges these signals to form the broadband signal. Compare the spectra of the input and output signals. The input and output spectra match very closely.

for i = 1:5000
    x = sum(sinewave(),2);
    y = channelizer(x);
    v = synthesizer(y);
    spectrumAnalyzer(x,v)
end

Channelizer with Complex Coefficients

Create a dsp.Channelizer object and set the LowpassCoefficients property to a vector of complex coefficients.

Complex Coefficients

Using firpm, determine the coefficients of a Park-McClellan's optimal equiripple FIR filter of order 30, and frequency and amplitude characteristics described by F = [0 0.2 0.4 1.0] and A = [1 1 0 0] vectors, respectively.

Create a complex version of these coefficients by multiplying with a complex exponential. The resultant frequency response is that of a bandpass filter at the specified frequency, in this case 0.4.

blowpass = firpm(30,[0 .2 .4 1],[1 1 0 0]);
N = length(blowpass)-1;
Fc = 0.4;
j = complex(0,1);
bbandpass = blowpass.*exp(j*Fc*pi*(0:N));

Channelizer

Create a dsp.Channelizer object with 4 frequency bands and set the Specification property to 'Coefficients'.

chann = dsp.Channelizer('NumFrequencyBands',4,'Specification',"Coefficients");

Pass the complex coefficients to the channelizer. The prototype filter is a bandpass filter with a center frequency of 0.4. The modulated versions of this filter appear with respect to the prototype filter and are wrapped around the frequency range [$-$Fs Fs].

chann.LowpassCoefficients = bbandpass

chann = 
  dsp.Channelizer with properties:

      NumFrequencyBands: 4
      OversamplingRatio: 1
          Specification: 'Coefficients'
    LowpassCoefficients: [1x31 double]

Visualize the frequency response of the channelizer.

fvtool(chann)

More About

Analysis Filter Bank

The analysis filter bank consists of a series of parallel bandpass filters that split an input broadband signal, x(n), into a series of narrow subbands. Each bandpass filter retains a different portion of the input signal. After the bandwidth is reduced by one of the bandpass filters, the signal is downsampled to a lower sampling rate commensurate with the new bandwidth.

Prototype Lowpass Filter

To implement the analysis filter bank efficiently, the channelizer uses a prototype lowpass filter. This filter has an impulse response of h[n], a normalized two-sided bandwidth of 2π/M, and a cutoff frequency of π/M. M is the number of frequency bands, that is, the branches of the analysis filter bank. This value corresponds to the FFT length that the filter bank uses. M can be high on the order of 2048 or more. The stopband attenuation determines the minimum level of interference (aliasing) from one frequency band to another. The passband ripple must be small so that the input signal is not distorted in the passband.

The prototype lowpass filter models the first branch of the filter bank. The other M – 1 branches are modeled by filters that are modulated versions of the prototype filter. The modulation factor is given by the following equation:

$$e^{-j{w}_{k}n},w_k=2\pi k/M,k=0,1,\mathrm{...},M-1$$

Using the Prototype Lowpass Filter

The transfer function of the modulated kth bandpass filter is given by:

$$H_k(z)=H_0(z{e}^{-j{w}_k}),w_k=2\pi k/M,k=1,2,\mathrm{...},M-1$$

This figure shows the frequency response of M filters.

To obtain the frequency response characteristics of the filter H_k(z), where k = 1,...,M-1, uniformly shift the frequency response of the prototype filter, H₀(z), by multiples of 2π/M. Each subband filter, H_k(z), {k = 1,..., M – 1}, is derived from the prototype filter.

Shift Narrow Subbands to Baseband

The frequency components in the input signal, x(n), are translated in frequency to baseband by multiplying x(n) with the complex exponentials, $$e^{-j{w}_{k}n},w_k=2\pi k/M,k=1,2,\mathrm{..},M-1$$ , where $$w_k=2\pi k/M$$, and $$k=1,2,\mathrm{...},M-1$$. The resulting product signals are passed through the lowpass filters, H₀(z). The output of the lowpass filter is relatively narrow in bandwidth. Downsample the signal commensurate with the new bandwidth. Choose a decimation factor, D ≤ M, where M is the number of branches of the analysis filter bank. When D < M, the channelizer is known as oversampled or non-maximally decimated channelizer.

The figure shows an analysis filter bank that uses the prototype lowpass filter.

y₁(n), y₂(n), ..., y_M-1(n) are narrow subband signals translated into baseband.

Algorithms

Polyphase Implementation

The analysis filter bank can be implemented efficiently using the polyphase structure. To derive the polyphase structure, start with the transfer function of the prototype lowpass filter:

$$H_0(z)=b_0+b_1{z}^{-1}+\mathrm{...}+b_N{z}^{-N}$$

N+1 is the length of the prototype filter.

You can rearrange this equation as follows:

$$H_0(z)=\begin{array}{c}\left(b_0+b_M{z}^{-M}+b_{2M}{z}^{-2M}+\mathrm{..}+b_{N-M+1}{z}^{-(N-M+1)}\right)+\\ z^{-1}\left(b_1+b_{M+1}{z}^{-M}+b_{2M+1}{z}^{-2M}+\mathrm{..}+b_{N-M+2}{z}^{-(N-M+1)}\right)+\\ \begin{array}{c}⋮\\ z^{-(M-1)}\left(b_{M-1}+b_{2M-1}{z}^{-M}+b_{3M-1}{z}^{-2M}+\mathrm{..}+b_N{z}^{-(N-M+1)}\right)\end{array}\end{array}$$

M is the number of polyphase components.

You can write this equation as:

$$H_0(z)=E_0(z^M)+z^{-1}{E}_1(z^M)+\mathrm{...}+z^{-(M-1)}{E}_{M-1}(z^M)$$

E₀(z^M), E₁(z^M), ..., E_M-1(z^M) are polyphase components of the prototype lowpass filter, H₀(z).

The other filters in the filter bank, H_k(z), where k = 1, ..., M-1, are modulated versions of this prototype filter.

You can write the transfer function of the kth modulated bandpass filter as $$H_k(z)=H_0(z{e}^{-j{w}_k})$$.

Replacing z with ze^-jw_k,

$$H_k(z)=h_0+h_1{e}^{-jwk}{z}^{-1}+h_2{e}^{-j2wk}{z}^{-2}\mathrm{...}+h_N{e}^{-jNwk}{z}^{-N}$$

N+1 is the length of the kth filter.

In polyphase form, the equation is as follows:

$$H_k(z)=\left[\begin{array}{ccccc}1& e^{-j{w}_k}& e^{-j2w_k}& \cdots & e^{-j(M-1)w_k}\end{array}\right]\left[\begin{array}{c}{E}_0(z^M)\\ z^{-1}{E}_1(z^M)\\ ⋮\\ z^{-(M-1)}{E}_{M-1}(z^M)\end{array}\right]$$

For all M channels in the filter bank, the MIMO transfer function, H(z), is given by:

$$H(z)=\left[\begin{array}{ccccc}1& 1& 1& \cdots & 1\\ 1& e^{-j{w}_1}& e^{-j2w_1}& \cdots & e^{-j(M-1)w_1}\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ 1& e^{-j{w}_{M-1}}& e^{-j2w_{M-1}}& \cdots & e^{-j(M-1)w_{M-1}}\end{array}\right]\left[\begin{array}{c}{E}_0(z^M)\\ z^{-1}{E}_1(z^M)\\ ⋮\\ z^{-(M-1)}{E}_{M-1}(z^M)\end{array}\right]$$

Maximally decimated channelizer (D = M)

When D = M, the channelizer is known as the maximally decimated channelizer or critically sampled channelizer.

Here is the multirate noble identity for decimation, assuming that D = M.

For illustration, consider the first branch of the filter bank that contains the lowpass filter.

Replace H₀(z) with its polyphase representation.

After applying the noble identity for decimation, you can replace the delays and the decimation factor with a commutator switch.

For all M channels in the filter bank, the MIMO transfer function, H(z), is given by:

$$H(z)=\left[\begin{array}{ccccc}1& 1& 1& \cdots & 1\\ 1& e^{-j{w}_1}& e^{-j2w_1}& \cdots & e^{-j(M-1)w_1}\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ 1& e^{-j{w}_{M-1}}& e^{-j2w_{M-1}}& \cdots & e^{-j(M-1)w_{M-1}}\end{array}\right]\left[\begin{array}{c}{E}_0(z)\\ E_1(z)\\ ⋮\\ E_{M-1}(z)\end{array}\right]$$

The matrix on the left is a discrete Fourier transform (DFT) matrix. With the DFT matrix, the efficient implementation of the lowpass prototype based filter bank looks like the following.

Non-maximally decimated channelizer (D < M)

When D < M, the channelizer is known as the non-maximally decimated channelizer or oversampled channelizer. In this configuration, the output sample rate is different from the channel spacing. In addition to that, the non-maximally decimated channelizers offer increased design freedom, but at the expense of increasing computational cost.

Here is the multirate noble identity for decimation, assuming that D < M.

For illustration, consider the first branch of the filter bank that contains the lowpass filter.

Replace H₀(z) with its polyphase representation.

After applying the noble identity for decimation, you can replace the delays and the decimation factor with a commutator switch.

For all M channels in the filter bank, the MIMO transfer function, H(z), is given by:

$$H(z)=\left[\begin{array}{ccccc}1& 1& 1& \cdots & 1\\ 1& e^{-j{w}_1}& e^{-j2w_1}& \cdots & e^{-j(M-1)w_1}\\ ⋮& ⋮& ⋮& \ddots & ⋮\\ 1& e^{-j{w}_{M-1}}& e^{-j2w_{M-1}}& \cdots & e^{-j(M-1)w_{M-1}}\end{array}\right]\left[\begin{array}{c}{E}_0(z^{M/D})\\ E_1(z^{M/D})\\ ⋮\\ E_{M-1}(z^{M/D})\end{array}\right]$$

The matrix on the left is a discrete Fourier transform (DFT) matrix.

The commutator switch in the case of the non-maximally decimated channelizer starts at the position (M/D) − 1 and progresses up the stack to branch 0. In the two-dimensional polyphase filter, new inputs samples are delivered to the first set of M/D addresses in the first column. The formal contents of these addresses are shifted to the next set of M/D addresses. The content in the last set of M/D addresses shifts to the first M/D addresses in the next column. All the samples in the two-dimensional filter undergo a serpentine shift. This is equivalent to performing a linear shift through the prototype one-dimensional filter prior to the polyphase partition.

For D = 2, here is how the data in the memory moves:

For every M/D samples that is fed as an input, the polyphase structure outputs M samples. This process increases the output sample rate from f_s/M to Df_s/M.

For more details, see [2].

After each D-point data sequence is delivered to the partitioned M-stage polyphase filter, the outputs of the M stages are computed and conditioned for delivery to the M-point FFT. The data shifting through the filter introduces frequency-dependent phase shift. To correct for this phase shift and alias all bands to DC, a circular shift buffer shown in Fig.30 of [2] is inserted after the polyphase filters and before the M-point FFT.

With the commutator switch followed by M-stage polyphase filter, circular shift buffer, and a DFT matrix, the efficient implementation of the lowpass prototype-based filter bank looks like:

References

[1] Harris, Fredric J, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004.

[2] Harris, F.J., Chris Dick, and Michael Rice. "Digital Receivers and Transmitters Using Polyphase Filter Banks for Wireless Communications." IEEE^® Transactions on Microwave Theory and Techniques. 51, no. 4 (2003).

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using MATLAB® Coder™.

Usage notes and limitations:

See System Objects in MATLAB Code Generation (MATLAB Coder).