msmucr
Member
The main difference is that I was only driving two channels and they are driving all 8. Someone commented that the bug in bass management may be the problem here and this would sort of point to that direction.
Amir,
thanks for the review.
Just what came to my mind about tilted noise spectrum you've measured.
To me this looks like rather typical issue with accumulated roundoff artifacts in recursive (IIR) filters.
Say for example if someone implements common direct form 2nd order HPF with single precision floats all the way through (input, coefficients, intermediate values for calculation, output product) in a straightforward way, then its FFT plot will be similar to what you've got (sans other distortion from DAC and analog circuitry of course).
In those cases the issue comes down recursive filter nature, where feedback imply use of previous values for calculation. With coefficients for low corner frequency and long decaying time domain response towards low frequencies of signal you are quickly in a situation, where algorithm has to deal with comparatively very small and very large float numbers. So despite of huge inherent range of floating point, there is relatively significant accumulated error, which shows on the plot as this tilted spectrum rising towards lows.
See my example attached pic with single prec HPF to 80 Hz and how it can look like on FFT.
There are general ways to push those artifacts well bellow any sane dynamic range of audio signals.. Most of high quality filters uses 64 bit double precision floats for critical parts of its algorithm, when used platform allows that. Also some filters are possible to redesign with different structure, which improves level of artifacts from calculations there.
Fixed point filter implementation and calculation is of course different in nature without "sliding" range of floats, but also in such case one has to deal with its fixed dynamic range for recursive calculations in terms of possible overflows and truncation, se even there it is advantageous to have more bits for numerical representation and calculation of critical parts.. say like in case of old Motorola/Freescale 56k with its 48bit double precision fixed point registers and associated 56 bit accumulators, those were commonly used for pro audio for high quality effects and crossovers. So ways are definitely there, if platforms allow and a designer wants to do that
I haven't seen this AV preamp, but isn't possible, there was some enabled bass management for measured channels...? Like some AV receivers has default 80 Hz HPF for front speakers, which are set to "small" at its setup. It should be apparent from its frequency response measurement.
If this was enabled, it could explain difference to the other measurement from Emotiva forum without that LF junk.
Michal