manisandher

I'll do some null-testing on the analogue outputs of the whole bunch of DACs that I seem to have accumulated across various systems. I suspect this might be quite involved, so will start a new thread once it's done. FWIW, I've found that switching to a multichannel DAC and a fully active speaker setup has had a far greater impact on sound quality than changing a filter, or sample rate, etc. ever could. Redbook sounds phenomenal here... or so say all my visitors 🙂. Mani.

One of MQA's claims was ringing due to filters (and not to the inevitable consequences of bandlimiting). But I see your point. Mani.

No, I think DeltaWave calculates nulls in the time domain, by looking at differences in levels and timing. But I'll let @pkane2001chime in here, if he wishes to. FWIW, the nulls in DeltaWave are very sensitive to different types of filters. Fast/steep linear-phase provide by far the best nulls in the audioband. Mani.

1kHz square wave sampled at 352.8kHz. Decimated down to 44.1k using steep/fast linear-phase filter. Upsampled back to 352.8k. Compared in DeltaWave: Correlated Null Depth = 180dBA (we're interested in audioband - we know they don't null above 22.05k) Good enough for me. Mani.

Maybe. But null-testing is the 'gold standard' IMO. If two files null to <-200dB, there's simply no getting around it.. they are essentially identical. Steep/fast linear-phase filters achieve this. [Of course, what happens in the UHF region is another matter 🙂.] Mani.

Agreed. You will only get the filter in a DAC to ring if there is clipping. It's the only way a non-bandlimited signal could get through, and pretty much what I've been saying in my last few posts. There's no reason that an anti-alias/decimation filter shouldn't work as well as a reconstruction filter. For example: I take a 24/352.8 white noise file at -2dBFS (so, lots of content right up to 176.4kHz). I decimate down to 24/44.1, using a steep linear-phase anti-alias filter. I upsample back to 24/352.8 and compare. They null to <-210dB in the audioband. Mani.

Well, if anything is being added, it sits at <-210dB. Mani.

Yes, and I believe this is totally irrelevant. There will be no ringing, because the input to the DAC will be bandlimited to fs/2. As mentioned a number of times now, my null testing shows that fast/steep linear-phase filters null to <-210dB in the passband, for any audio/test file (necessarily bandlimited to fs/2) I throw at them, provided there is no clipping. Mani.

What are the "issues" using a fast/steep linear-phase brickwall filter (other than perhaps latency)? Mani.

We established a while ago in this thread that there are is an infinite spectrum with copies of the baseband at each side of multiplies of fs. I'm talking about the baseband itself. Mani.

I'm refering to this: A digital audio file will never contain them, as it will always be bandlimited. What might expose the reconstruction filter's response is clipping, but this is easily dealt with. Mani.

I'm not talking about what comes out of a DAC, I'm talking about what goes in. It is IMPOSSIBLE for a non-bandlimited signal to enter a DAC. Mani.

@bogior @Tsarnik, can you explain how you would get a non-bandlimited signal into a DAC?

Thanks Nicholas. Yes, I can see the aliasing pretty clearly. As you said, there's no way of removing it and keeping the 'real' content. It's a shame, as there's no excuse for it being there. I suppose the argument is that it's at a low level, and at a very high frequency. But still... Thanks again. Mani.

Could you show an example of a spectrogram where it's "straightforward" to identify aliased content please? Mani.