Pre-ringing does not exist in linear phase well designed brick wall filtering for produced consumer music playback. ( a well designed filter should start just at or above 20 KHz - i.e. beyond the audible range) The only way you could get ringing out of playback system with linear phase is bad filter within audible band or a square wave or impulse in the signal - this is a step change at the sampling rate. This is not music. This is a test signal. Properly produced Music does not have signals like this.
Minimum phase creates huge problems of phase accuracy - it may be necessary if using sharp filters within the audible band where pre-ringing can occur, as you might run into when in a studio while mixing. It should never be used for playback where preservation of phase information is necessary to maintain fidelity.
The ability to add convolution filters is a nice cost effective solution for major room space issues but it
is no longer high fidelity (faithful to the recording)
and room acoustics should be addressed if fidelity is desired.
The last couple of points depends on what is the shared accepted meaning of “high fidelity” - so folks will beg to differ for sure. However my points about Pre-ringing and minimum phase are scientific in origin if you view playback as trying to faithfully replicate the recording.
Yup. Agreed. Not for everyone for sure but when getting into the realm of experimenting with DSD and using high resolution files - perhaps one is already well down the rabbit hole!
Yes, it does, already in the source content, due to the band-limiting introduced by the ADC (or production) decimation anti-alias filter. I have shown this in practice already long time ago, several times. It is easy to produce with acoustic instruments, good microphone (that is flat to 20+ kHz) and transient signals, like soprano glockenspiel.
In addition, lot of source content contains other types of errors as well, which some can be fixed at the playback time. (most ADCs and production tools are pretty poor)
Just for fun, you can monitor “Apod” analysis counter in HQPlayer while playing back 44.1/48k content and check how much of your listening content trips it to figures over 10 during a single track.
In addition, yes it does. Especially thanks to loudness wars. Lot of content is heavily clipped causing almost square waves with flat tops.
No, it is more faithful to the recording, since the acoustic signal you are hearing is closer to the correct one.
And HQPlayer has filters with less than 0.000000001 dB ripple, and over 300 dB stop-band attenuation. And can run proper filters to final output rate (1024x for example).
ES9038 is typical example of resource constrained implementation cutting way too many corners.
It’s filters are pretty poor, and in addition it cannot run proper filters more than 8x and after that is is very rudimentary oversampling with 3rd order IIR (minimum phase!) filter. Their only filter with 0.002 dB pass-band ripple has just -120 dB attenuation (equivalent to 20 bit precision) and is leaky, reaching maximum attenuation only by 0.55 fs. Rest of their filters have even poorer responses and attenuations around 100 dB or less, so barely matching 16-bit precision. In addition their modulator has various distortion issues due to cutting those corners. But you probably know and can detect it’s sonic fingerprint yourself too.
With the filter you mentioned, it’s reconstruction accuracy is equivalent to about 13.4 bits.
I guess you would be happy with this kind of HQPlayer’s linear phase filter for 44.1k content?
You get lower distortion and better reconstruction accuracy (roughly 30 dB better).
Not sure where to start. You seem to have a good understanding of tap filtering and math but a very poor understanding of physics. My background is from physics so context is always important to know if a number is meaningful or not.
Firstly, 0.002 dB is the “ripple” in passband on the outstanding ES9038 - nothing poor about this specification. The reconstruction accuracy is still excellent and to say it is only 13 bit is wrong. The slight ripple simply means slight differences in amplitude with frequency for audio in the passband. Since this is three orders of magnitude less than a typical speaker it is completely negligible from a physics perspective.
Secondly, if your source file is a victim of loudness wars then the sound is compromised whatever you do. Physics says it doesn’t matter much in this case as the file is garbage whatever you do - even so any pre-ringing in this instance will occur outside the audible range. - again from a physics perspective not worth worrying about.
Thirdly, pre-ringing from AtoD filtering in the recording stage can NOT be removed by a minimum phase filter in playback. Minimum phase filters are not time machines - they can’t go back to fix the recording. The pre-ringing from AtoD will be there whatever filter is used, and should be inaudible (above 20KHz) and if audible then within the audio band then cannot be removed unless a brick wall filter of lower corner frequency is used (this will also affect the highest audible frequencies of real instruments)
Finally, how processing gets you “closer to the correct one” - well I don’t know what that is - a bit of hand waving there to me - my specific concern is preservation of critical phase information that minimum phase filters tend to introduce. Pre-ringing isn’t a concern unless it affects the in band audio and if it does then the audio is compromised anyway.
I agree your filter from HQ player is absolutely superb mathematically. I just have trouble when I put on my physics hat and ask so what? All this additional precision is irrelevant. I would argue that most hardware chips and DAC implementations don’t bother with going to such huge levels of precision because it is processing intensive, completely inaudible compared to an adequate filter (audio physics) and just adds audio latency with no tangible benefit.
I think you might want to go back to physics books
That reconstruction accuracy is combination of the digital filters plus the typical analog output filter. Error in the output signal is at about -80 dB level.
Yes it can, it is easy to demonstrate and do. Apodizing filters can replace the original impulse response with another one.
Some of the things can be corrected, or dealt better with. And same goes for correcting for less than great ADCs.
This is something I let each one choose when it comes to filters in HQPlayer. There are more linear phase filters than minimum-phase, plus some intermediate phase ones as well.
But just remember that your analog path, including DAC output stage contains minimum phase filters. Typical net result is at least some tens of degrees at 20 kHz, before loudspeakers.
If your loudspeaker use passive cross-over you may get less error, but typically some time coherency issues still between drivers. You can see this from the step/impulse response. If it has analog active cross-over then the phase errors are typically similar or more. If it has simple digital IIR based cross-over, then it will be minimum-phase.
If it’s convolution based, then it can be correct and we get back to the digital room correction. Since you could then as well include the room corrections in the convolution filters.
Many HQPlayer users run multi-way cross-overs in HQPlayer with or without room correction. Many modern room correction systems can also correct phase/timing response of the loudspeaker in addition to correcting it’s magnitude response. Thanks to what you can do with convolution filters in general.
And yet pretty much all DAC chips on the market have various different filters to choose from. Included for no reason? Have you listened yourself? I don’t have trouble hearing differences between different filters.
And we didn’t even touch modulator implementations yet.
In player software’s we have at least couple of seconds worth of RAM buffer when reading from disk or network, and some more for audio output, so as long as latency of filters is less than one second, it is not really noticeable.
Typical room correction filter is from 64k taps to 256k taps. So these can exceed one second.
I’m pretty OK on that front too. I started with underwater acoustics / passive sonar systems plus radars in the early 90’s. HQPlayer development was started along that stuff in 1998. Later I moved on from the military systems to other areas.
I would claim that I still know enough especially about physics of underwater acoustics that there are not so many people on this globe knowing more.
From that work and listening a lot I also know how sensitive instruments trained human ears can be. Picking up things from seemingly pure noise. You can also train your hearing.
I understand you worked on Sonar in the Finnish navy Jussi.
Is it true that @ipeverywhere should stop listening to EDM (electronic dance music) because it ruins his high frequency hearing and he will never make a good sonar operator ?
Many many studios today use a combo of room acoustics design, room treatments and digital room correction
If you want to be faithful to the recording (recording? the mix? the mastering? it is unclear what you mean…) then the ideal is obviously to replicate that as much as possible
Don’t be afraid of convolution - it might get you closer to the studio sound !
I see now what formed your bias and subsequent misapplication of the power of convolution/deconvolution filtering with apodizing filters. Typically, wavelet deconvolution would be something you do in elastic wave processing like seismology. Apodizing filters are typically used to smooth out the impact of bandwidth limiting (the box car as it is called in time series analysis). This typically reduces “ringing” from the sharp cut offs. The misunderstanding here is that audio signals are akin to impulse responses. They are not. Proper music production requires bandwidth limiting the analog signal PRIOR to A to D. Any impulse responses in a music file are errors/garbage as an result of digital manipulation of the data after A to D. If there are errors/garbage in the file then we can simply stop there and admit there is nothing that can be done to restore the damage.
Unfortunately the physics of a bandwidth limited signal means that in the real world you can’t process the result and get back the information lost - not without an impossible infinite precision as well as knowledge of the exact filter used. Without these details it is a shot in the dark and likely to add spurious noise artifacts.
In sonar/radar just as in room correction or reflection/refraction seismology, the deconvolution depends on taking a measurement of the system response to a known signal - an impulse or a chirp signal are common test signals. Typically a first reflection or a major system multiple reflection may be used in lieu of a measured test signal. Processing out multiples using a wavelet deconvolution filter can make processed images cleaner by removing ghost artifacts.
With a measurement of system response or first reflection then one can go one step better and try to correct or deconvolve to a desired wavelet. However care must be taken as the desired wavelet should not add frequency content that wasn’t there originally as this just creates spurious noise. Room correction can be used to reduce room reflections and modes in this manner (based on test measurements).
Unfortunately with a final music production there will be a multitude of microphones recording at multiple locations all filtered differently and mixed together. In the case of a musical recording, there is no single deconvolution that could fix the entire final file - this is impossible and a very much more complex and entirely different situation from a sonar source pinging a 3D space (where you could record and analyze first reflections).
My point is that the physical world is limited. The inverse problem - I.e. correcting the deficiencies in the A to D studio ADC on a mixed recording of a multitude of bandwidth limited microphone recordings is not possible in general and could only be undertaken in a forensic individual specific effort and really with only the multi-channel original recordings of said individual recordings (prior to a mix)
So a lengthy tap HQPlayer filter can mathematically reject -300 db out of band. So what? This is so far beyond the real physical world that as impressive as this sounds it can’t physically sound better over an average good filter like in the ES 9038 chip option for fast linear phase.
Ah but you might say - folks all confirm that with a multitude of DACs there is an audible difference when using extreme filters. And you would be right. I would say all those DACs with audible differences must therefore have poor implementation of their internal filter or offer a choice of filters of which some are objectionably leaky - like the Holo May KTE NOS option where there is only a very minor slow filter - obviously that highly rated DAC is designed to necessitate a pre-filtered processed digital signal to operate correctly in that mode. From cursory reading it seems that the Holo May DAC in NOS mode is best fed a very high rate DSD input in order to push noise up into much higher frequencies where the gentle filter will have enough attenuation or where the ultrasonic noise won’t pass through the amps and speakers.
Good point! Many small and home studios have adopted a room correction approach but it is not without issues. The top studios eschew manipulation like this because preserving phase information helps make things sound natural/real on the final product. Strong DSP manipulation such as room eq generally necessitates degradation of phase information (adds group delay). Studio engineers will use small nearfield bridge monitors (less room effect) and mains (more room effect) alternatively depending on tasks… the very best studios spend a fortune on room acoustic treatments and construction. Due to speaker crossover phase issues some engineers rely on 2 or 3 different speakers to see how things sounds. Most engineers learn to work with a small number of industry widely accepted “reference” speakers that are known and understood and “translate” well.
This is getting boring since you obviously don’t know anything about the topic.
It is not misapplication, you can also find bunch of scientific papers on the topic if you are interested.
Your beloved ES9038PRO also has an apodizing filter. Too bad it is as compromised as other ES9038PRO filters are. (just like it’s modulator)
That will happen always when you have band limiting. Band limiting filter will always leave it’s fingerprint on the data. Only way to avoid this is to use high enough sampling rate that it doesn’t require band-limiting. Which is the case with DSD ADC for example.
Wrong.
I’m not talking about getting removed information back. I’m talking about removing the additional error signal from the base signal.
Now you are mixing fixing the source content vs fixing the playback chain errors.
It is.
Yes it can, because this is much more complex than you may seem to realize.
They are audible also on all my ES9038PRO DACs as well. I have some tens of different DACs I use for R&D and QA purposes. Too bad the ESS modulator has it’s own sonic fingerprint that you cannot get rid of completely.
Yes, much better, since we don’t need to pay for the resource constrained / compromised digital filters and modulators that are in the ESS chip. All the necessary processing can happen at higher quality along with the playback in the already existing computer. And the DAC manufacturer can actually put money where it matters - in the actual D/A conversion stage and analog stages.
As result, May can perform much better.
Different filters sound different for multitude of reasons. Even if they are linear-phase with equal pass-band ripple and stop-band attenuation. I leave it for you as home work to figure out why, it is pretty obvious if you are familiar with the topic.
Whenever I listen to my ESS based DACs (rarely), I just bypass it’s internal filters by feeding those with 705.6/768k PCM or DSD512. DSD is shortest path through those DACs.
Wrong, it can correct both phase response of the system, such as DAC and amplifiers, in addition to correcting phase response of the loudspeakers. With convolution based correction, you can adjust magnitude/frequency, phase/frequency and time independently.
Here are some of the correction filter generation tools I recommend:
Yes you can make adjustments to kingdom come provided you know what the original audio is supposed to be or was before it hit the AtoD in the studio etc. etc. Your statements are nearly all theoretical. A lot of what you claim is totally impractical, given the lack of information contained in a consumer audio file (which is all that consumers often have available).
In the real world we are limited with the limited data available and therefore what is practical.
ES9038 is just an example of a practical solution. It measures quite well. Why would I love it? It is an example of ‘adequate’ when regarding the fast linear phase filter. Inadequate if you regard the other filters (poor performance in this regard)
Digital filtering through HQplayer is a very powerful additional tool. I fully support what HQplayer are doing especially when it comes to some of the many boutique DACs that indeed are poor or offer poor solutions as an option. However, you do your followers a disservice when you make such bold claims about fixing. To fix something you need to know exactly what is wrong and by how much!
MQA made the mistake of making exaggerated claims regarding what they could achieve with apodizing filters, I would advise not to follow them in this regard.
(Above I use poor performance from a scientific measured basis and not from a listener perspective. Tubes sound great but they often don’t measure as well. Many boutique DACs serve the purpose of producing a delightful sound that listeners enjoy and that is a great thing)
What @jussi_laako has done, while never needing to, is to show actual results from his software and how the software influences the waveform. He’s spent countless hours comparing what he does against others in the industry like Rob Watts. Tons of discussion on theory and we, as users, just need to look it up. How that influences what we hear has always been an exercise for the user and plenty of users agree with the methodology put forth within HQP. Other users prefer Watts. Other users prefer Ted Smith. There are plenty of choices to choose from and especially if you want to dive deep. Your argument, that this isn’t audible, and the existing filters are “adequate” seems misplaced against the mountain of comments from those of us who actually listen.
Where MQA went really wrong is they claimed to fix the errors on a per-DAC basis. They magically had some way of correcting ADC errors in the DAC through licensing and certification and tuning of each MQA DAC. That’s just bunk. Everyone knew that was just bunk. This isn’t the claim of HQP. The claim, of how I understand HQP, is that a) there are always errors b) pick what errors you want to correct but nothing is perfect. That does require the user have some basic understanding of what is being corrected and how that may improve / not improve sound quality. Notes are provided in the HQP user manual to give people a chance to find something that works for them.
There is always the argument that speakers are terrible at reproducing sound so why are we worrying so much about the precision of the voltage at the DAC? I’ll quote Mike Moffat again “audio precision handjob b*t is just flea farts while jet engines are going by”.
However, I hear a difference. That’s good enough for me. I don’t need to understand why. I just want to listen to my music and smile about it. I just happen to have a background where the “why” can be part of the hobby. I can find “entertainment” in comparing what Laako, Watts, and Smith are trying to do. It’s easiest for me to get access to HQP than the other brands. So, here we are. Happy listening.
I’ll second that - I’ve been so impressed by a year with the RME ADI-2 DAC FS (ESS) for my headphone rig, I’ve bought the ADI-2 Pro FS R BE for my main speaker rig. I did play around with DSD Direct and upsampling by Roon but preferred upsampled PCM. I may explore HQP one day, which was part of my reason for choosing the Pro (DSD Direct).
Agreed. I never actually said everything was inaudible. Lots of subtle differences are quite audible especially in the context of poor DAC designs and use of leaky filters. MQA is an audible detriment to the original audio quality likely because of Minimum Phase filtering.
I fully support HQplayer. What an excellent and incredibly flexible tool! Used correctly it is extremely powerful.
I only question certain specific claims above that I feel are rather exaggerated - like the idea that a band pass brick wall filter with -120 dB suppression of out of band being poor and all DAC chips being poor. I counter that they are designed to a price point and designers made choices that they believe makes a great DAC chip at a good price - the designers stop short of additional complexity to process huge tap filters because this is far beyond diminishing returns (inaudible improvements in the context of even the latest electronics and equipment).
I do not regard the latest ESS audio chips as crap. I don’t like many of their inadequate, distorting and leaky standard filters but used correctly with fast linear phase filtering this is an excellent design at a good price supremely capable and sufficient for the modern era.
HQ Player, and a Holo Audio May KTE are my current setup. Tried a bunch of other DACs of all sorts with and without HQ Player. No question that even casually listening the Holo and HQP combo is the most satisfying, impressive and easiest to listen to of what I have tried.
Many have been impressed. NOS mode with HQplayer upsampling to high rate DSD seems the preferred. Is that your preference?
Upsampling to high rate DSD in HQplayer may allow the lack of a steep brick wall filter in the NOS mode to be effective by pushing noise high up in ultra-high frequencies.
Not having a steep brick wall filter in NOS mode at 44.1 KHz is an odd implementation from a scientific point of view - it will inevitably suffer from aliasing.
