Converting a Matlab filter-design to Matlab code and obtaining filter coefficients

Central problem (Abstract):

Using the Matlab filter-designer results in direct-form II transposed coefficients, which i need. However, generating Matlab code from the filter-designer app itself does not yield these coefficients as function-returns.

• use function to create automated filterbanks
• Store all coefficients in C-header file
• Use coefficients in MCUs with ARM CMSIS DSP library
• Filter type: 3 section biquad IIR Butterworth

Filter designer: These are the coefficients i need.

Code:

Generated code:

function Hd = gen_iir_osix_bp(Fc1, Fc2, Fs)

% EDITED AUTOGEN FROM FILTER DESIGNER

%GEN_IIR_OSIX_BP Returns a discrete-time filter object.

% MATLAB Code
% Generated by MATLAB(R) 9.10 and Signal Processing Toolbox 8.6.
% Generated on: 18-Nov-2021 12:27:19

% Butterworth Bandpass filter designed using FDESIGN.BANDPASS.
N   = 6;      % Order

% Construct an FDESIGN object and call its BUTTER method.
h  = fdesign.bandpass('N,F3dB1,F3dB2', N, Fc1, Fc2, Fs);
Hd = design(h, 'butter');

% [EOF]

Own code:

fs = 48000;     % [Hz] desired sampling rate of device
fmin = 20;      % [Hz] minimal desired low-end
fmax = 20480;   % [Hz] minimal desired high-end

oct_bw_ratio = 1/3; %  desired octave-split

n_octs = log2(fmax/fmin);               %  number of octaves from fmin to fmax
n_bands = ceil(n_octs/oct_bw_ratio);    %  number of octave divisions from fmin to fmax

fc = fmin*2.^((0:n_bands) * oct_bw_ratio);  % [Hz] center frequencies.
% "(0:n_bands)" means to iterate from 0 to n_bands
% and store result in vector
fl = fc*2^(-oct_bw_ratio/2);                % lower cutoffs @ -3dB (required by filter-designer)
fu = fc*2^(+oct_bw_ratio/2);                % upper cutoffs @ -3dB (required by filter-designer)

% test filter with uppermost band
filter = gen_iir_osix_bp(fl(31), fu(31), fs);
display(filter.sosMatrix);

Output:

1.0000         0   -1.0000    1.0000    1.2094    0.6087
1.0000         0   -1.0000    1.0000    1.8842    0.9026
1.0000         0   -1.0000    1.0000    1.4369    0.5143

Not the same coefficients. Which ones represent what exactly?

• The output of filter.sosMatrix is already a series of three biquad IIR filter and I believe it is the coefficients that you need. You said "not the same coefficients" but I don't know compared with what? If you mean CMSIS uses 5 coefficients but MATLAB gives you 6, check it out here Nov 22 '21 at 4:08

I'm not quite sure what you're asking, but it looks like you have three second order sections. The

[1.0000 0 -1.0000]

are the numerators and the remaining numbers on each line are the denominators.

Does that help?

If I do

[b,a] = sos2tf(filter.sosMatrix)
freqz(b,a)

then I get this plot, which looks right to me, based on your plot above. I've not used the SOS much before, but if I dig into it:

display(filter)

filter =

FilterStructure: 'Direct-Form II, Second-Order Sections'
Arithmetic: 'double'
sosMatrix: [3x6 double]
ScaleValues: [0.268829511895073;0.268829511895073;0.242859077404354;1]
OptimizeScaleValues: true
PersistentMemory: false

it looks like the sosMatrix component doesn't scale things properly, and the scale is captured in the ScaleValues component of the designed filter.

If I calculate the attenuation using those scale values:

20*log10(0.268829511895073*0.268829511895073*0.242859077404354)

ans =

-35.1138

then I get something close to your noted 40dB gain.

• The coefficients seem so "unusual" to me: if i plug them into a visualizer tool then i don't see an expected bandpass form. I get that the coefficients represent 3 different filters of second order in series, but what is the filter designer giving me then? Nov 21 '21 at 17:04
• @jake_asks_short_questions This seems all correct to me. What's the problem again?
– Peter K.
Nov 22 '21 at 0:40
• It seems i just don't know enough about math, i was missing the connection between the output of the filter-designer and the return of the discrete call of the code generated by the designer. But the function "sos2tf" establishes that connection for me Nov 22 '21 at 8:15
• @jake_asks_short_questions ok! Great! Let me know if there’s anything I can add.
– Peter K.
Nov 22 '21 at 11:26
• I'll edit the main question to reference your answer and add my own answer below, as i think this post gives a good example of a filter use-case in the designer and its counterpart in code, thank you for your contribution Nov 22 '21 at 16:26