Channel analyzer NG: autocorrelation corrections (2): corrected FFT aliasing

plugins/channelrx/chanalyzerng/chanalyzerng.cpp

@@ -52,8 +52,7 @@ ChannelAnalyzerNG::ChannelAnalyzerNG(DeviceSourceAPI *deviceAPI) :
 	m_inputFrequencyOffset = 0;
 	SSBFilter = new fftfilt(m_settings.m_lowCutoff / m_inputSampleRate, m_settings.m_bandwidth / m_inputSampleRate, ssbFftLen);
 	DSBFilter = new fftfilt(m_settings.m_bandwidth / m_inputSampleRate, 2*ssbFftLen);
-	m_corr = new fftcorr(4*ssbFftLen);
-	//m_pll.computeCoefficients(0.05f, 0.707f, 1000.0f); // bandwidth, damping factor, loop gain
+	m_corr = new fftcorr(8*ssbFftLen); // 8k for 4k effective samples
 	m_pll.computeCoefficients(0.002f, 0.5f, 10.0f); // bandwidth, damping factor, loop gain
 
 	applyChannelSettings(m_inputSampleRate, m_inputFrequencyOffset, true);

sdrbase/dsp/fftcorr.cpp

@@ -43,7 +43,7 @@ void fftcorr::init_fft()
     outptr = 0;
 }
 
-fftcorr::fftcorr(int len) : flen(len)
+fftcorr::fftcorr(int len) : flen(len), flen2(len>>1)
 {
     init_fft();
 }
@@ -66,7 +66,7 @@ int fftcorr::run(const cmplx& inA, const cmplx* inB, cmplx **out)
         dataB[inptrB++] = *inB;
     }
 
-    if (inptrA < flen) {
+    if (inptrA < flen2) {
         return 0;
     }
 
@@ -83,6 +83,7 @@ int fftcorr::run(const cmplx& inA, const cmplx* inB, cmplx **out)
     }
 
     std::transform(dataA, dataA+flen, dataBj, dataP, [](const cmplx& a, const cmplx& b) -> cmplx { return a*b; });
+
     fftA->InverseComplexFFT(dataP);
 
     std::fill(dataA, dataA+flen, 0);
@@ -95,7 +96,7 @@ int fftcorr::run(const cmplx& inA, const cmplx* inB, cmplx **out)
     }
 
     *out = dataP;
-    return flen;
+    return flen2;
 }
 
 const fftcorr::cmplx& fftcorr::run(const cmplx& inA, const cmplx* inB)

sdrbase/dsp/fftcorr.h

@@ -41,6 +41,7 @@ public:
 private:
     void init_fft();
     int flen;  //!< FFT length
+    int flen2; //!< half FFT length
     g_fft<float> *fftA;
     g_fft<float> *fftB;
     cmplx *dataA;  // from A input