• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Assessing DSP quality

dweeeeb2

Active Member
Joined
Feb 10, 2023
Messages
288
Likes
275
Location
Melbourne
I’m curious, what’s people take on DSP “quality”. I don’t see it mentioned very often so was wondering if people generally consider all “graphic” equalizers the same, all PEQ equalizers the same etc.

If I was to apply a +2dB PEQ at 2kHz with a 0.7Q would I expect the same result on all PEQ’s.
 
If I was to apply a +2dB PEQ at 2kHz with a 0.7Q would I expect the same result on all PEQ’s.

My take:

DSP is all math, applied to the source digits as they pass through the filter buffer.

The results are repeatable.
 
Last edited:
@RayDunzl is correct for PEQ filters but traditional "graphic equalizers" as found in many vintage rack systems *edit and even VST plugins had/have no standard I'm aware of regarding bandwidth coverage. The key part of this is the Q part of the equation.
 
There are several differences you can expect from one EQ implementation to another.

There are two main types of Q interpretation: constant Q and proportional Q, you will see the difference explained page 4 here. This is the most important source of audible difference you can expect, and one that should certainly not be ignored.

Another (lesser) difference will be the behavior in high frequencies, as cramping will happen when approaching Nyquist with many implementations. So the response in HF might vary depending on the sampling frequency as well as from one implementation to another depending on the way this cramping phenomenon is handled. You can read more about this here.

These two differences are linear distortion, but you also have non linear distortion that will happen with IIR due to quantization errors, especially in the low frequencies, and especially with 32 bit implementations (most hardware ones). This might look innocuous at first but these errors do add up, and there are worse case scenarios, even with 64 bit implementations.
 
Last edited:
@pos LOL, I was going to mention some of this but didn't want to "muddy the waters" too much but was thinking of the various rePhase videos and tutorials I've seen.

I've been using your software a lot lately dialing in my recent woofer replacements in my active 3-way XO. THANKS!
 
  • Like
Reactions: pos
There are several differences you can expect from one EQ implementation to another.

There are two main types of Q interpretation: constant Q and proportional Q, you will see the difference explained page 4 here. This is the most important source of audible difference you can expect, and one that should certainly not be ignored.

Another (lesser) difference will be the behavior in high frequencies, as cramping will happen when approaching Nyquist with many implementations. So the response in HF might vary depending on the sampling frequency as well as from one implementation to another depending on the way this cramping phenomenon is handled. You can read more about this here.

These two differences are linear distortion, but you also have non linear distortion that will happen with IIR due to quantization errors, especially in the low frequencies, and especially with 32 bit implementations (most hardware ones). This might look innocuous at first but these errors do add up, and there are worse case scenarios, even with 64 bit implementations.
Some interesting links there, thank you.
 
I’m curious, what’s people take on DSP “quality”. I don’t see it mentioned very often so was wondering if people generally consider all “graphic” equalizers the same, all PEQ equalizers the same etc.

If I was to apply a +2dB PEQ at 2kHz with a 0.7Q would I expect the same result on all PEQ’s.
You mention at the start "DSP quality", then mention "graphic equalizers". So I'm not sure if you are including traditional hardware devices in your question. Traditional analogue graphic equalizers (e.g. 1970s, 1980s) were not digital, so were not DSP (Digital SP). They used analogue low- high- and band-pass filters. Mixing desks of that era often had much more advanced Parametric EQ allowing centre frequency and Q to be adjusted - these were uncommon in domestic HiFi.

Modern mathematics-based DSP "graphic equalizers" and the more advanced PEQ behave quite differently from these original analogue designs. In almost all cases, they measure much better and suffer fewer downsides.
 
There are several differences you can expect from one EQ implementation to another.

There are two main types of Q interpretation: constant Q and proportional Q, you will see the difference explained page 4 here. This is the most important source of audible difference you can expect, and one that should certainly not be ignored.

Another (lesser) difference will be the behavior in high frequencies, as cramping will happen when approaching Nyquist with many implementations. So the response in HF might vary depending on the sampling frequency as well as from one implementation to another depending on the way this cramping phenomenon is handled. You can read more about this here.

These two differences are linear distortion, but you also have non linear distortion that will happen with IIR due to quantization errors, especially in the low frequencies, and especially with 32 bit implementations (most hardware ones). This might look innocuous at first but these errors do add up, and there are worse case scenarios, even with 64 bit implementations.
Thanks for the links!
Can we involve in this quality the ability to not induce noise when DSP down low as we see with some devices like the Bluesound Node,MiniDSP,Topping,Wiim,etc?
In one case the comparison with RME filters for example showed some clear superiority from the RME side.
 
There are several differences you can expect from one EQ implementation to another.
Hello Pos.
What rePhase adjustments can you recommend for 30-300Hz. 3 room mode filters based on measuring REW. For 44100/16.
Q constant or Q proportionnel? Linear phase or minimal?
Thank you.
 
...
There are two main types of Q interpretation: constant Q and proportional Q, you will see the difference explained page 4 here. This is the most important source of audible difference you can expect, and one that should certainly not be ignored.

Another (lesser) difference will be the behavior in high frequencies, as cramping will happen when approaching Nyquist with many implementations. So the response in HF might vary depending on the sampling frequency as well as from one implementation to another depending on the way this cramping phenomenon is handled. You can read more about this here.
....
Thank you, great info. Does this suggest that using "generic eq" in REW to generate PEQ's could be causing unreliable results when using a miniDSP HTx (The HTx is not listed in the REW EQ options)? EDIT: just updated to latest version and found Flex with Dirac is an option now

Modern mathematics-based DSP "graphic equalizers" and the more advanced PEQ behave quite differently from these original analogue designs. In almost all cases, they measure much better and suffer fewer downsides.
Yes, I was referring to the digital versions.

Thanks for the links!
Can we involve in this quality the ability to not induce noise when DSP down low as we see with some devices like the Bluesound Node,MiniDSP,Topping,Wiim,etc?
In one case the comparison with RME filters for example showed some clear superiority from the RME side.
So, its sounds like not all DSP EQ's are considered equal. Is this "noise" considered audible?
 
Last edited:
@dweeeeb2 Check out

Edit: Not quite what you're looking for, I now realize..
 
So, its sounds like not all DSP EQ's are considered equal. Is this "noise" considered audible?
I would guess no but not for sure.
In this thread which is about digital measurements all the way (so no analog sins messes it further as for example analog in - DSP - analog out) we can see some pretty good hit in some,some less in others and some no penalty at all.
It's a combination of filter implementation and hardware's capability.But they are not all of equal quality for sure.

Edit: @staticV3 got me first.
 
What rePhase adjustments can you recommend for 30-300Hz. 3 room mode filters based on measuring REW. For 44100/16.
Q constant or Q proportionnel? Linear phase or minimal?
For sound reproduction minimum-phase EQ should be preferred (not talking about crossovers here).
You can use either constant or proportional Q as long as you get to your target at the end. The only difference between these two is the way the bandwidth is calculated depending on the gain: they are equivalent when this difference is accounted for.
There are other differences when it comes to shelving filters for example (the slopes, and more importantly the position of the cutoff frequency), but to be clear these differences only matter when you try to replicate a given correction based on settings. When designing a correction based on measurements this is a non issue (as long as you have enough EQ points and types, of course).

Can we involve in this quality the ability to not induce noise when DSP down low as we see with some devices like the Bluesound Node,MiniDSP,Topping,Wiim,etc?
In one case the comparison with RME filters for example showed some clear superiority from the RME side.
Yes, this is quantization noise.
 
A little quantization noise and dither experiment:

 
Back
Top Bottom