Hello, I am seeing ~ 1% changes in total power in a 1 MHz PFB spectrometer channel when I apply a test tone 20 dB above the noise in 1 MHz many channels away. This corresponds to ~ 40dB leakage rejection between channels, which seems lower than I would have thought for a 2-tap, 2048 point Hamming window, for which I would expect ~120 dB rejection in voltage, so 60dB in power. I have tried this experiment in a PFB both without rounding and with round to even enabled, with the same result. Robert Jarnot suggested that this could be due to intermodulation products from the test tone going through the discretized transfer function of the ADC. Have other people experienced this problem? Is there any way around it? It seems like this severely limits the PFB's usefulness.
Thanks, Glenn
