I've been listening to the N017 today and I think there is too much post ringing at the moment. I think it needs to be about midway between the current balance and linear. It's got the "turn up the reverb to 11" thing happening at the moment, which I'm not really keen on. If I can get it wound back to 5 or 6 it would be ideal
Post up what you are trying to do with MATLAB and we should be able kick start the process. I'm not great wiz with it, but can do basics.
The second test is altering the stop band frequency to start at 22.05kHz, rather than 24kHz. The stop band attenuation is set to 110dB, so a little worse than the stock filter.
The steeper filter will alter the sound somewhat but in theory the haze should be there, but harshness should be somewhat reduced.
Just did a really quick check and "So What" seems smoother and the trumpet looses a brittle edge with the cut-off dropped back to nyquist.
The steeper filter will alter the sound somewhat but in theory the haze should be there, but harshness should be somewhat reduced.
Just did a really quick check and "So What" seems smoother and the trumpet looses a brittle edge with the cut-off dropped back to nyquist.
Pass Band Ripple
One of the variables I forgot to factor in was pass band ripple.
I'd been changing this without really thinking.
The stock filter has a pass band ripple spec of 0.01dB which - based on Gerzon's EQ sound article - will be large enough to impart colouration. I've using a ripple spec of around 0.0001-0.00005 dB as my ideal, but sometimes fall back to as low as 0.001 if necessary.
I've come across posts by Miska (developer of HQPlayer) suggesting his preferred ripple spec is 0.000006dB maximum. So the obvious thing to try was altering the default SK441 configuration by the pass band spec to 0.0000006dB and leaving everything else to default.
This test filter is attached as 1021filt_SK441_lowPBRip.skr
Lowering the pass band ripple seems to take some of the edge off the stock filter, to my ear such as it is.
From there the next logical step was to increase stop band attenuation while leaving the pass band and stop band edge frequencies as per the stock filter. I picked 130dB stop band attenuation - for no real reason - and keep the 0.0000006dB pass band ripple constraint.
Test filter attached as 1021filt_HQ_130.skr
This is quite listenable, and doesn't sound too forward or harsh to my ears.
Increasing the stop band attenuation to 150dB with a stop band slope of 24dB/Oct seems to improve the recovery the ambience of the recording venue.
Attached as 1021filt_HQ_150_spb.skr
It really should read ssb rather than spb but you know what I mean...
Next step is to push the stop band attenuation to what I think is the useful limit for the DAM - 170dB. The stop band slope has been increased to 48db/Oct.
Attached as 1021filt_HQ_170_spb8.skr
It's getting marginal but I think there is an improvement to ambience and detail recovery. I could be imagining it of course.
One of the variables I forgot to factor in was pass band ripple.
I'd been changing this without really thinking.
The stock filter has a pass band ripple spec of 0.01dB which - based on Gerzon's EQ sound article - will be large enough to impart colouration. I've using a ripple spec of around 0.0001-0.00005 dB as my ideal, but sometimes fall back to as low as 0.001 if necessary.
I've come across posts by Miska (developer of HQPlayer) suggesting his preferred ripple spec is 0.000006dB maximum. So the obvious thing to try was altering the default SK441 configuration by the pass band spec to 0.0000006dB and leaving everything else to default.
This test filter is attached as 1021filt_SK441_lowPBRip.skr
Lowering the pass band ripple seems to take some of the edge off the stock filter, to my ear such as it is.
From there the next logical step was to increase stop band attenuation while leaving the pass band and stop band edge frequencies as per the stock filter. I picked 130dB stop band attenuation - for no real reason - and keep the 0.0000006dB pass band ripple constraint.
Test filter attached as 1021filt_HQ_130.skr
This is quite listenable, and doesn't sound too forward or harsh to my ears.
Increasing the stop band attenuation to 150dB with a stop band slope of 24dB/Oct seems to improve the recovery the ambience of the recording venue.
Attached as 1021filt_HQ_150_spb.skr
It really should read ssb rather than spb but you know what I mean...
Next step is to push the stop band attenuation to what I think is the useful limit for the DAM - 170dB. The stop band slope has been increased to 48db/Oct.
Attached as 1021filt_HQ_170_spb8.skr
It's getting marginal but I think there is an improvement to ambience and detail recovery. I could be imagining it of course.
More illustrating numbers for the passband ripple you get perhaps if you look at how much the signal differs from the goal, i.e at the number x, if 1-x is you ripple. This then looks like the numbers for the stopp band, the SNR, etc.
E.g. your 0.0000006dB are "1 - (-143dB)", so correspond to 24 bit resolution.
Or the other way round from a 16-bit source you can expect due to quantization a ripple of
"1- (-96dB)" ~ 0.0001dB (have we not seen that number somewhere )
With the DAM you can not control more than 28 bit, so
"1- (-168dB)" ~ 0.00000003dB
...
E.g. your 0.0000006dB are "1 - (-143dB)", so correspond to 24 bit resolution.
Or the other way round from a 16-bit source you can expect due to quantization a ripple of
"1- (-96dB)" ~ 0.0001dB (have we not seen that number somewhere
)With the DAM you can not control more than 28 bit, so
"1- (-168dB)" ~ 0.00000003dB
...
Last edited:
More illustrating numbers for the passband ripple you get perhaps if you look at how much the signal differs from the goal, i.e at the number x, if 1-x is you ripple. This then looks like the numbers for the stopp band, the SNR, etc.
E.g. your 0.0000006dB are 1 - 143dB, so correspond to 24 bit resolution.
Or the other way round from a 16-bit source you can expect due to quantization a ripple of
1- 96dB ~ 0.0001dB (have we not seen that number somewhere
With the DAM you can not control more than 28 bit, so
1- 168dB ~ 0.00000003dB
...
OK, that makes sense. Is there any point in "over designing"?
I'd ended up at 170dB stop band, and 0.00000002dB pass band ripple.
Stepping back to 168dB and 0.00000003dB saves maybe 20 taps in the configs above, so if there is no advantage I'll put the taps to better use.
I would say no. You get these values (in theory) on the output only with a full scale signal. Musik almost never is. If you use the digital volume control you lose loose a lot of bits of resolution. So with the 168dB you are already "over designing" for the real application.
More important is that FIR2 is of the same quality (in the FIR1 passband). Due to the design algorithem maxima and minima of the ripple are correlated with fractions of fs. I.e. there are good chances that some maxima (minima) of FIR1 and FIR2 graph coincide (the ripple dBs then are added, i.e. you get at least the magnitude of the worst).
Last edited:
EQHQ_V1
I've done a few tweaks to the last HQ_170_spb8 filter and this is now pretty much a complete and consistently designed filter.
Whether it sounds any good I'll leave up to your ears...
The changes from HQ_170 are:
- 0.0000003dB pass band ripple constraint - all filters
- 168dB stop band attenuation - all filters
- transition band tightened from 5kHz to 3.69kHz, now -0.008dB @ 20kHz. (44.1kHz)
- pass band edge raised from 19kHz to 19.5kHz (44.1kHz)
- New 48kHz filter which takes advantage of extra 2kHz before Nyquist cf 44.1kHz.
- New equiripple filters with transition band from 20kHz to 90% of Nyquist for 88.2, 96, 176.4, 192kHz sample rates.
- Tweaked FIR2 filter for 352.8kHz sample rates. This is currently used for 384kHz also.
I've done a few tweaks to the last HQ_170_spb8 filter and this is now pretty much a complete and consistently designed filter.
Whether it sounds any good I'll leave up to your ears...
The changes from HQ_170 are:
- 0.0000003dB pass band ripple constraint - all filters
- 168dB stop band attenuation - all filters
- transition band tightened from 5kHz to 3.69kHz, now -0.008dB @ 20kHz. (44.1kHz)
- pass band edge raised from 19kHz to 19.5kHz (44.1kHz)
- New 48kHz filter which takes advantage of extra 2kHz before Nyquist cf 44.1kHz.
- New equiripple filters with transition band from 20kHz to 90% of Nyquist for 88.2, 96, 176.4, 192kHz sample rates.
- Tweaked FIR2 filter for 352.8kHz sample rates. This is currently used for 384kHz also.
I think so. The EQHQv1 is dead flat to over 20kHz.
Actually did a comparison between 16/44.1 and 24/192 copies of Miles Davis' So What and was pretty pleased that there was little to pick, apart from the level of ambience recovery like the sense of room on the bass, between the two.
Last edited:
I've had a listen to the C128_100 as I hadn't heard it for a while.
Personally I think it highlights the two problems that I've been trying to address with more recent filters:
- artificial brightness that results from excessive imaging above nyquist
- apparent haze in recordings and loss of detail resulting from inadequate stop band attenuation.
Both are apparent on the C128_100 to my ears.
Personally I think it highlights the two problems that I've been trying to address with more recent filters:
- artificial brightness that results from excessive imaging above nyquist
- apparent haze in recordings and loss of detail resulting from inadequate stop band attenuation.
Both are apparent on the C128_100 to my ears.
Throughout the months I've been trying many filters but ended up going back to NewNOS, however EQHQ v1 is a keeper by far. Its more revealing but doesn't have the brightness of the other filters and keeps it at a similar natural level as NewNOS while keeping the detail of the highs.
I'm glad you didn't take a break on filters just yet. Great work!.
I'm glad you didn't take a break on filters just yet. Great work!.
Thanks.
Went back and listened to the NewNOS cf the EQHQ_v1. It's surprising how close they are in terms of weight and richness.
I'd listened to the wrong filter... C128_100bp is the full precision version, and sounds better. My main criticism is that it lacks weight and body. There is a lightness and "bounce" to the C128 that is nice but I'd rather the sense of solidity that is missing.
I ended very soon up with listening to music insted comparing with other filters, I think this is a good sign. Thanks also for also the higher fs filters.
Ha!
Fell into the trap of confusing imaging distortion with "detail"...
The EQHQ filters were getting brighter as the transition band was shifted further up, and the stop band moved further past Nyquist,
"oh good, more air and better detail"...
bzzzzt, wrong answer!
Lead on to the rocks by the siren song of "fools detail".
The best solution to this problem will have to wait for the firmware update assuming we get more taps... until then you have the choice of an adopising filter, - in that it's fully attenuated at 22050, and -3dB at 19.9kHz - or a slightly less strict filter (quasiapo) that fully attenuates at 22500kHz and is -3dB at 20.37kHz.
Fell into the trap of confusing imaging distortion with "detail"...
The EQHQ filters were getting brighter as the transition band was shifted further up, and the stop band moved further past Nyquist,
"oh good, more air and better detail"...
bzzzzt, wrong answer!
Lead on to the rocks by the siren song of "fools detail".
The best solution to this problem will have to wait for the firmware update assuming we get more taps... until then you have the choice of an adopising filter, - in that it's fully attenuated at 22050, and -3dB at 19.9kHz - or a slightly less strict filter (quasiapo) that fully attenuates at 22500kHz and is -3dB at 20.37kHz.
Last edited:
I know what you mean.
I've got to the point where I feel like there isn't much point messing around further until the firmware update comes out.
Have a listen to the two above. They don't really lose out in terms of detail, and are smoother for the removed/reduced imaging.
Once the firmware update comes out I should be able to move the pass band edge up by about 1kHz and still zero out by 22.05khz
Last edited: