Convolution with Null filters are audible

Leong_Kin_Foong1 · October 27, 2018, 10:56am

Switch Convolution in Roon on or off with the Null filter in place, I can hear a difference. My system is pretty revealing and it’s there - sounds better without convolution. Can some people with high end systems please just listen and see if they can hear it too? No need to measure anything, just listen. I’m not technical enough to be able to propose measurement tests etc.

Or do I need my brain examined?!

If there is a way to measure the difference, of course it’s welcomed for discussion. Thanks Marco for working on this to date.

Larry_Post · October 28, 2018, 3:47pm

Can you post or host a set of null filters?

Leong_Kin_Foong1 · October 29, 2018, 3:18am

I did. It’s attached to my post above near the top. A single null WAV file for Roon convolution plug-in.

alec_eiffel · October 29, 2018, 6:52am

Yep. Pretty easy to generate with Rephase as well if you want to play with various sample rates, filter lengths…

ogs · October 29, 2018, 2:12pm

My weekend did not allow time to do the testing of the NULL filter, sorry about that! I did not listen to music at all… I’ll see if I can do it during the week. I really want to do this!

Larry_Post · October 29, 2018, 4:55pm

My apologies I didn’t see it. I’ve had a listen with and without and do detect a bit of less dynamic sound. There may be a slight attenuation with but it’s not much, the loss of dynamics is more apparent to my ears.

For my room, the slight loss of dynamics is worth it for the correction but I agree there shouldn’t be any difference with a null filter.

Marco_de_Jonge · October 30, 2018, 10:59pm

If you agree that there is a lack of ‘dynamics’, then you have a noble task to do.
If you were asked to describe this feature in great detail : how would you describe it ?

Or even better I should ask : how would your experience, translate into a measurable quantity ?
Without such a thing, I’m afraid that nobody can solve anything for you…

Just a quick hint : we are already pretty certain, that it hasn’t much to do with the scientific definition of dynamics. But what is it, then ?

Another hint-ish something : convolution by itself, is not capable of dynamic compression. Even if you wanted it to do that. So, what else could be happening ?

Leong_Kin_Foong1 · November 1, 2018, 4:24am

I’m not sure how much more detail I can offer to describe the drop in dynamics when using Roon convolution. It’s like there is less emotion in the music and less musicality. It’s makes me less engaged when listening to it. It’s like a drop back to a lesser set of hifi equipment. As for translating this to a measurable quantity, apart from my earlier attempts to use an iPhone app where I found a clear result, it’s beyond me.

At this time, I am not sure if convolution is the way to go for solving room nodes - perhaps it may be better to stick with physical changes to the room and traps although this is tougher to do. My original though was that it helped a lot but over time and critical listening I am changing my mind about the benefits of Roon convolution for higher levels of HiFi. For Mid-Fi, it’s fine I suppose.

evand · November 1, 2018, 8:03am

No amount of DRC on its own is ever going to give you a better result than room treatment in conjunction with DSP. Ideally DRC should be what’s done after room treatment is in place.

Leong_Kin_Foong1 · November 1, 2018, 8:06am

That’s what I used to feel - doing both. But now I am more inclined to believe that no amount of convolution is acceptable. It just messes up the sound - even with null filters in place. The technology or technique being used by Roon isn’t working for me yet.

evand · November 1, 2018, 8:14am

I’d suggest it’s with DRC, not specific to Roon’s implementation thereof. I also find volume leveling robs music if its dynamics so I don’t use it unless I just have Roon radio DJ’ing whilst I’m doing something else and music is just a backdrop.

Marco_de_Jonge · November 1, 2018, 11:05pm

I saw it’s been said already. But yes, this is always the case. Treating the actual acoustical problems, is always better than trying to fix them electronically. I know this is not what you like to hear, but it’s just the unconvenient truth…

I have already written in a PM, that any change in the low-end must be a WHOLE LOT larger, than the loss in detail/dynamics that you’re experiencing. Whichever you prefer is a personal decision, and I respect that.

On a personal note : I’m still curious to know, what causes your experience. The only logical explanation to me, is that you hear the effects of double bit depth conversion including dithering. Nothing more seems to be happening.

Just to confirm/deny : can you try to listen for the difference of any random other DSP operation, ? For instance, this one on/off, it should hardly do anything :

Audiogeek · November 1, 2018, 11:38pm

Interesting discussion. I was told for a long time that Roon upsampling sounds just as good as HQP upsampling by others using the same front end as I have. However, subjectively I always preferred HQP - because of the convolution function.
I would suggest the op try convolution in hqp with hqp integration in Roon.
For bass issues due to room modes etc., convolution is a very effective solution. At least with correcting peaks and high shelves etc. Dips still require speaker/listener positioning or more bass sources to correct.
Have no idea how to measure these differences in convolution between hqp and roon, or to give a plausible explanation.
Thierry/HAF always produce better sounding filters than I have been able to generate in Acourate or REW.

Leong_Kin_Foong1 · November 2, 2018, 2:36am

Yes, will try this suggestion and report shortly about trying other DSP. And yes Thierry has been very helpful and fast in this response. He made both Roon and HQPlayer filters for me.

They sound roughly the same but it’s not a simple swap to compare sounds for me which require a cable disconnect/reconnect so critical comparisons are harder. Also I don’t have null filters for HQP, only for Roon to try.

ogs · November 3, 2018, 9:51pm

I have finally found time to listen to the NULL convolution.

I used a 16bit track from Ana Caram’s Chesky album Rio After Dark, track 4 - Viola Fora De Mode for listening.

directly with DSP disabled

2. with convolution and the NULL file

DSP enabled, headroom at -1dB, clipping indicator disabled - all other DSP off, +1dB volume on amp (to have the same loudness as 1&2)

DSP enabled, headroom at -1dB, clipping indicator disabled - convolution with NULL file enabled, again +1dB volume on amp

4.PNG377×615 31.9 KB

There is a difference between 1 and 2. I’d describe 2 as having a little lack of detail and immediacy in the over tone structure compared to 1. I.e. it is less dynamic.

I can not detect a difference between 2, 3 and 4. Note that 1 is different from the others in that Roon appears to be sending the 16bit file directly to microRendu while in 2,3 and 4 data is transferred as 32bit. I do not think this makes a difference in sound, but I can not be certain of course.
I believe it is the dithering that makes the difference as 2,3 and 4 all go down from 64bit float and dithering is likely on for all these. That means that convolution (and the other pure mathematical DSP functions) in it self is probably transparent. A different dithering algorithm might be better here or maybe the possibility to select the dithering to use depending on what equipment is used for playback.

All this is listening only, no measurements where done. My system is Roon core (360) on a i5 NUC running Debian, microRendu with LPS1, Mutec MC3 + USB, Devialet 250CI and modified MA-201 speakers.

Leong_Kin_Foong1 · November 4, 2018, 5:49am

Ah that’s a good logical test of DSP. And more data points that confirm that I did hear a difference with Null filters. There are at least
3 people in this thread that heard something now.

This concept of dithering that may be the cause of the drop in dynamics and engagement, perhaps someone from Roon can think on this and have a look? Can it be improved upon?

ogs · November 4, 2018, 9:54am

Dithering is a good thing and is considered useful when reducing bit-depth or re-sampling. I think the effect of dithering depends on what is done to the signal in the next stage(s). If the DAC also does dithering (or up-sampling) this may be ‘too much’. As far as I know Roon has no possibility to know what is done to the stream at the next stages so dithering is applied. Then my suggestion of selecting the dithering algorithm to use or even disabling dithering all together on certain specific setups may make sense.

ogs · November 10, 2018, 8:29am

I am a bit disappointed as I hoped @brian would add a comment here. Maybe I got it all wrong? Only he knows how this actually works in Roon - and whether selectable dithering is possible.

brian · November 10, 2018, 9:22pm

Sorry! I only get notified if you flag @brian. This site is long past the point where I read every single thread like back in 2015

There is a reduction in bit-depth whenever Roon performs any processing, because Roon’s processing uses a 64bit format, and DACs, in general, support 24 or 32 bits. So somewhere inside of Roon, we need to get back from that 64bit format to 16-32bits.

I don’t think disabling dither makes sense–the bit-depth reduction demands it, regardless of what it happening next. It’s important to realize that if you don’t dither, and simply truncate the bits of a twos-complement value, the output signal is technically invalid because it includes a DC offset.

The truncation/dither + adjustment for the DC offset all happen together as one step, and really can’t be separated without sacrificing correctness. You could not get a technically correct result by truncating the numbers to, say, 24bit, and then trusting some downstream process to dither, since there is already information lost. To do this correctly, you add the dither to the higher-precision value, compensate for the DC offset in that higher-precision domain, and then truncate.

Allowing for dither algorithm selection is a fine idea. Maybe we will do it at some point in the future.

I find it fairly unlikely that you are hearing a dither at the 24th bit. Dither is a neutral noise pattern, and at that position, it is at least 40dB beneath the noise floor of your listening environment. Especially when starting with 16bit content and doing 32bit output, there is just too much mathematical precision there to suspect it.

When dither is implemented badly, it does not create a character that I would consider consistent with " lack of detail and immediacy".

It is far more likely, in my view, that one of the following is creating the difference:

The pair of volume adjustments, in particular, the +1dB on the amp (since the -1dB in DSP is information preserving in the above signal paths due to the large increase in mathematical precision).
Doubling the data rate to the DAC by sending 32bits instead of 16bits. Some people feel that choice of transfer format changes the sound of some DACs.

I don’t think any of the above contradicts your subjective test results. Just some thoughts…

Leong_Kin_Foong1 · November 11, 2018, 3:07am

I’ve been told that sending using a 32bit scan of a cd sounds better than a 16bit rip. In the past I’ve tried a dozen cd scans at 32 and 16 bits and can’t hear much difference if any.