Fix chebyshev band pass and band stop returning all NaN

[adityasingh2400]

DataFilter.perform_bandpass and DataFilter.perform_bandstop return an array that is entirely NaN whenever the filter type is CHEBYSHEV_TYPE_1 or CHEBYSHEV_TYPE_1_ZERO_PHASE. The exit code is still STATUS_OK, so nothing is raised and the NaN silently poisons whatever runs next, band powers, PSDs, the ML metrics. Butterworth and Bessel are fine, and Chebyshev low pass and high pass are fine, so only these two entry points are affected.

The cause is a parameter slot mismatch against the bundled DSPFilters library. ChebyshevI::Design::LowPass and HighPass are built on TypeI, which takes four params in the order sample rate, order, cutoff, rippleDb, so the ripple really does belong in slot three there. ChebyshevI::Design::BandPass and BandStop are built on TypeII, which takes five, sample rate, order, center frequency, band width, rippleDb, as declared by getParamInfo_3 = defaultBandwidthHzParam and getParamInfo_4 = defaultRippleDbParam in third_party/DSPFilters/include/DspFilters/ChebyshevI.h. Both perform_bandpass and perform_bandstop wrote params[3] = ripple for every Chebyshev type. On the band designs that overwrote the band width with the ripple value and left slot four unwritten, and since Dsp::Params is a plain struct with no constructor and clear is never called, the design was computed from a bogus band width and an indeterminate ripple.

The fix writes the ripple into params[4] for the band designs, leaves the band width in params[3], and calls params.clear () first so an unwritten slot can never be read as garbage again. Chebyshev low pass and high pass are untouched. This is why the bug survived so long: every Chebyshev call site in the repository, python_package/examples/tests/signal_filtering.py, cpp_package/examples/signal_processing/src/signal_filtering.cpp and julia_package/brainflow/test/signal_filtering.jl, uses low pass or high pass only. No example or test ever calls Chebyshev band pass or band stop.

I built on macOS arm64 with cmake .. && make -j8 and drove the Python binding against the freshly built libDataHandler. Before the change, a 4096 sample signal at 250 Hz through perform_bandpass(..., CHEBYSHEV_TYPE_1, 1.0) came back 4096 of 4096 NaN, and the same for perform_bandstop, while Butterworth and Bessel were clean. After the change, using a sum of sine probes at 1, 10, 25, 60 and 90 Hz, a 5 to 30 Hz Chebyshev band pass gives per tone gains of 0.0019, 0.9475, 0.8948, 0.0049 and 0.0002, closely tracking the Butterworth reference of 0.0012, 0.9991, 0.9544, 0.0176 and 0.0007. A 20 to 30 Hz band stop notches the 25 Hz tone to 0.0005 and leaves the others above 0.89.

Two further checks confirm the specific slots. The band width is now honoured rather than being replaced by the ripple value: sweeping the band as 5 to 10, 5 to 30 and 5 to 60 Hz moves the 25 Hz gain from 0.0012 to 0.8948 to 0.9272 and the 60 Hz gain from 0.0003 to 0.0049 to 0.8890. The ripple now reaches the design and behaves monotonically: at 0.1, 1.0 and 3.0 dB the 10 Hz passband gain is 0.9943, 0.9475 and 0.8361, which is the expected deepening of passband ripple. The zero phase variant is also NaN free and correct.

For regressions I ran the synthetic board examples CI already runs, denoising.py, signal_filtering.py, transforms.py, downsampling.py, band_power.py, band_power_all.py, windowing.py, ica.py and csp.py, and all pass. clang-format with the repository .clang-format reports no change anywhere in the file. There is no assertion based test surface for the signal processing functions, the existing coverage is demo scripts that plot rather than assert, so I could not extend one. I am happy to add a Chebyshev band pass case to the examples if you would like the path covered in CI.

[Andrey1994]

The parameter-slot correction matches the bundled DSPFilters band-design parameter order, and params.clear() is appropriate defensive initialization.

Please add a focused regression case exercising both CHEBYSHEV_TYPE_1 bandpass and bandstop. At minimum, the test should assert that all output samples are finite; ideally it should also check basic passband and stopband attenuation. The existing signal-filtering examples do not exercise these combinations, which is how this path could regress without CI noticing.

[Andrey1994]

Sorry, that was codex and autoreview, change looks fine, thanks

[adityasingh2400]

Added python_package/examples/tests/chebyshev_band_filters.py and wired it into run_unix.yml.

It drives a 10 Hz plus 50 Hz signal through both CHEBYSHEV_TYPE_1 bandpass and bandstop with a 5 to 15 Hz band, asserts every output sample is finite, and checks attenuation per tone: the kept tone stays above 0.8 and the rejected one falls below 0.05. Measured values are 0.94 and 0.0015 for bandpass, mirrored for bandstop.

On master it fails at the first assertion with bandpass produced non-finite samples.

[adityasingh2400]

I missed your follow-up before posting that, sorry. I had started on the tests when the first comment came in and did not re-check the thread before pushing.

The test and the CI line are a separate commit, so if you would rather merge this as the one-line fix you already reviewed, say so and I will drop it. If you want to keep it, nothing further is needed.