Convolution sign fix

[ik1xpv]

A possible correction of convolution sign bug and gain correction. :-)

[howard0su]

Good catch. Oscar Steila <[email protected]>于2021年1月12日 周二下午7:30写道：

***@***.**** commented on this pull request. ------------------------------ In Core/fft_mt_r2iq.cpp <#151 (comment)>: > @@ -235,11 +235,11 @@ void fft_mt_r2iq::Init(float gain, int16_t **buffers, float** obuffers) // Bw *= 0.8f; // easily visualize Kaiser filter's response KaiserWindow(halfFft / 4 + 1, Astop, relPass * Bw / 128.0f, relStop * Bw / 128.0f, pht); - float gainadj = gain / sqrtf(2.0f) * 2048.0f / (float)FFTN_R_ADC; // reference is FFTN_R_ADC == 2048 + float gainadj = gain * 2048.0f / (float)FFTN_R_ADC; // reference is FFTN_R_ADC == 2048 I placed pht[t] both in pfilterht[][0] and pfilterht[][1]. This gives a real signal with sqrt2 gain, that is not present if we use pht[t] in pfilterht[][0] and pfilterht[][1] = 0. The index position of pfilterht[] was wrong too it is now [halfFft-1-t]. There was many errors compensating each other :-)) — You are receiving this because your review was requested. Reply to this email directly, view it on GitHub <#151 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAF3GRAEKKIZWZX72M77MSDSZQXEPANCNFSM4V65P7HQ> .

-- -Howard

[hayguen]

pfilterht[halfFft-1-t][0] = gainadj * pht[t]; looks similar to what i have in pffft

gainadj : here we should have a test, that verifies full scale (carrier in passband) at input produces exactly full scale at output - without any dBm correction depending on receiver model and preamp. also see #140
In the extIO and other tools, the dBm correction should also be deactivatable easily .. to be able to pass the correction value the user or SDR program separately.

looks good for me .. without having tested in detail.
but think, that Howard should contain the reversing of time domain filter directly in his pull request #148

[ik1xpv]

@howard,
@hayati,
please be free to merge it if of some use after #148 or close.