If the graphs where in db it would look a lot better, and be comparable to others graphs.
-
Welcome to ASR. 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!
DSP is bad!
- Thread starter boxerfan88
- Start date
MAB
Major Contributor
The cheapest graphic equalizer from 1980:
Costs more in present value (~$275) than a standalone DSP unit in 2025.
The old Radio Shack unit has audible distortion, can only do a few channels of EQ, can't be used as a crossover to make active speakers, completely sucks in comparison. And no it can't do much room and speaker correction with the set of filters it has. Of course there are more competent Graphic EQ, they are incredibly costly compared to DSP.
I'll leave aside that computers have free DSP applications available. Perhaps some people don't have or use computers.
48kHz is Q-SYS only sample rate. And there is a 16,384 tap limit per channel. I used to use all 16k taps on every output channel just for ease in setting fixed delays.
Till my processing guru showed me what I was leaving on the table with long filters where they aren't needed.
I'm finding 6k taps is the most I use anymore, for the sub with xover around 100Hz. Sub EQ's and hpf are IIR / min phase.
Then 2k taps for a 300Hz xover, 1k for a 750-900Hz xover, and 512 taps @4000Hz.
Those are numbers that work easily on my synergy multi-ways. About half those numbers also work for each xover, albeit with more care.
trying not to go too far off topic but
..... for the sub crossover I know you like to use Linear Phase filters so do you use pure "FIR Linear Phase filters" or "FIR Linearized IIR filters" and why?
No - graphic equalizers are useless for room correction. They can give you an adjustment to the overall frequency response to match your personal preference - that is all.
They are unable to target particular peaks or dips in your room response - nor are they able to introduce delays to (for example) integrate subs. They can do nothing that will improve impulse response.
Etc Etc.
OP
boxerfan88
Major Contributor
- Joined
- Oct 23, 2018
- Messages
- 1,335
- Likes
- 1,629
- Thread Starter
- #87
Would you be able to post the EQ banks/graph for VAR & ERB?Oh no a new rabbit hole to go down
See below, first graph is step response for my "old" correction filter generated in REW using VAR smoothing and 96 KHz to just above Schroeder. (I used to upsample everything to 96 KHz so I could use the same FIR room correction filter for all content but stopped that after I realized the resampling was causing digital clipping issues).
The second graph is my "new" light handed correction filter generated by REW from the same measurement and same target but using ERB smoothing and 44.1 KHz. I definitely prefer listening to the "light handed" correction even though if I measure the corrected response at the listening position it is considerably "less smooth" looking.
I am not experienced in correlating "looking at a step response and correlating it to what I hear" but just looking at the graph of my "old" filter it looks scary. I am starting to think "over doing" DSP is a very real and common thing.
I’m curious how see what the differences are…
OP
boxerfan88
Major Contributor
- Joined
- Oct 23, 2018
- Messages
- 1,335
- Likes
- 1,629
- Thread Starter
- #88
I’m trying to understand the workflow. If a min-phase FIR filter generated by REW is 32ktap long, and the coefficient values are not zero throughout 32ktap, how do we trim it to let’s say 8ktap?
By some special procedure? Or we just chop off at the 8k coefficients limit, and discard the rest?
I guess for linear phase, we need to center the peak, then chop off both ends?
Keith_W
Major Contributor
Can you explain what you are leaving on the table with long filters, apart from latency?
See below, top one is "Psychoacoustic Smoothing" (I thought it was ERB but was wrong), bottom one is "Var Smoothing". Done on the same measurement with the same REW filter settings. Pretty big difference!
BillG
Major Contributor
I personally don't believe that about it. DSP can accomplish much more than manipulating frequency response. In the past, I've used it for room modeling, tube amp modeling, loudness compensation, compression, limiting, up mixing, down mixing, cross feeding channels, reverb, echo, stereo expansion, stereo widening and ambisonics. However, I currently use very few of those functions and keep it simple with a touch of corrective EQ and loudness compensation.
OP
boxerfan88
Major Contributor
- Joined
- Oct 23, 2018
- Messages
- 1,335
- Likes
- 1,629
- Thread Starter
- #92
Oh wow, 18 filters vs. 10 filters. Also the Q is not so narrow in the psychoacoustic EQ. I should try it. Thanks for sharing.
trying not to go too far off topic but
I use straight FIR linear phase filters for several reasons.
First is that my FIR generators automatically build a FIR filter to give flat response thru the passband and match the xover curves desired.
It's nearly braindead simple, to get near perfect acoustic measurements.
Whereas using IIR is an arduous process at best, even with auto EQ's like REWs.
Here's an example of mid range drivers in a syn/MEH horn. Example works same for all sections..
First is the raw response.
Let's say I want to use this driver section from 300Hz to 900Hz, with 60dB/oct linear-phase xovers.
All I have to do is specify the xovers freq and type, the number of taps, smoothing, desired window, and then adjust output level via Peak Norm to get to 0dB.
Takes only a few minutes. I then get this acoustic result, with linear-phase.
If I use IIR to to try to get even close to the above acoustic response, it's gonna take a long time, and a lot of PEQ's, shelves, and playing with crossovers.
And then if I want to phase-linearize, yikes....one more step
Why the heck would I do it in two steps, one of which....the IIR step....is plain painful to do. ???
Second reason, is phase-linearization is valid for an IIR crossove, , It gives the same result as a simple lin-phase xover. If someone uses an inverse all-pass via FIR, like via rePhase, that's pretty valid (but again I question why do it in two steps).
However, that identity only holds true in one-dimension electric space.
If they do more than simply insert the specific inverse-all pass to phase-linearize the IIR crossover....if they take an acoustic measurement and then phase-linearize that measurement, they are going beyond what's valid in 1D space, and are correcting xover phase to a particular spot. Odds of that hit or miss linearization are much worse over polar space, than having true complementary linear-phase crossovers in place.
I've seen a number of folks try to phase-linearize an entire speaker that is already tuned either IIR active, or passive. They take an overall speaker measurement, and apply global FIR to linearize phase. Bogus. Like just said, only the electrical xovers, with an exact inverse all pass applied for each, are valid phase corrections. Whatever minimum-phase rolloffs exist in the speaker's measurement will also get phase linearized....a mistake imo.
Hope all that made sense.
Regarding DSP correction and phase issues it does not sound very consistent. But there is one aspect in this article that we missed to discuss so far:
That is certainly true, and one aspect of DSP room correction that is frequently overlooked. While it is pretty effective with problems that are affecting solely tonality, correcting the outcome of reflections causing localization, imaging or decay issues, is close to impossible.
It will help for me to point to the measurements I just posted to levimax. Please look at the raw impulse response at the top, and note all the continued oscillations past impulse peak. My room/environment continuing to talk into mic.
Then in the processed acoustic reponse above, after applying the 16,384 tap FIR filter, note how the acoustic impulse has cleaned up.
Here is that 16k tap FIR filter's electrical impulse. It's clear to see it is addressing the post peak oscillations .
Ok, now I'm going to apply a 1024 tap FIR filter to the raw response.
Here's the measured acoustic response with filter in place.
We can see magnitude response has a bit of ripple compared to perfect using 16k taps, and little correction occurs much past impulse peak.
Looking at the 1024k tap electrical impulse, it's clear the filter makes little attempt at time domain corrections well past impulse peak.
So for a little more mag ripple, much less heavy-handed time domain correction occurs, with fewer taps.
Like said earlier, my theory is the time domain overcorrection is akin to some kind of residual noise always being applied to signal. But I could be full of shit...
I also want to explore windowing the measurement, to reduce the time window the FIR filter tries to correct, and hopefully put more taps back into play to get back frequency resolution.
Another, hope that all made sense.
Last edited:
I use FIR filter generators where for linear-phase you simply specify the number of taps and sample rate you want to use.
I don't know how REW handles FIR filter length now. I remember a number of years ago trying REW as a FIR generator, and using IR Windows to make the necessary FIR size.
Worked then, maybe still does.
While we're on the topic of FIR crossovers, I'd like to highlight some specific features of FIR crossovers.
I'm sure many are familiar with the concept of pre-ringing in FIR filters. The longer the FIR filter, the longer the transient response in the digital filter. The image below shows a recording of a 2.5 kHz electrical signal at the output of an FIR crossover set to 2.5 kHz and 512 taps long. As we can clearly see from the measurement results, the HF and MF drivers receive out-of-phase signals during the transient response, and these signals cancel each other out when summed.
By analyzing the measurement results, we can identify a number of requirements that must be met when setting up a speaker system.
First, to ensure uniform cancellation of out-of-phase signals during the transient response, it's necessary to have the same delay from the HF and MF drivers at the listening position. Simply put, it's crucial to align the acoustic centers of the speakers.
Second, to ensure uniform subtraction of out-of-phase signals during the transient, it's necessary to have the same SPL from both speakers at the listening position within the frequency range of the speakers.
Third, at the end of the transient, there's a signal greater than the 2.5 kHz signal amplitude. If you don't provide "gain headroom," the amplifier may clip and introduce unwanted harmonics into the output.
An FIR filter is a very convenient feature, but to avoid making mistakes and achieve the desired result, you need to consider a number of specific operating characteristics of this type of filter.
I'm sure many are familiar with the concept of pre-ringing in FIR filters. The longer the FIR filter, the longer the transient response in the digital filter. The image below shows a recording of a 2.5 kHz electrical signal at the output of an FIR crossover set to 2.5 kHz and 512 taps long. As we can clearly see from the measurement results, the HF and MF drivers receive out-of-phase signals during the transient response, and these signals cancel each other out when summed.
By analyzing the measurement results, we can identify a number of requirements that must be met when setting up a speaker system.
First, to ensure uniform cancellation of out-of-phase signals during the transient response, it's necessary to have the same delay from the HF and MF drivers at the listening position. Simply put, it's crucial to align the acoustic centers of the speakers.
Second, to ensure uniform subtraction of out-of-phase signals during the transient, it's necessary to have the same SPL from both speakers at the listening position within the frequency range of the speakers.
Third, at the end of the transient, there's a signal greater than the 2.5 kHz signal amplitude. If you don't provide "gain headroom," the amplifier may clip and introduce unwanted harmonics into the output.
An FIR filter is a very convenient feature, but to avoid making mistakes and achieve the desired result, you need to consider a number of specific operating characteristics of this type of filter.
Just note that when you move your head up or down, the path differences change, and the summing will no longer be perfect. This is not just the case with direct sound. It's way worse for reflected sound, especially floor and ceiling bounce.
Yes, it's possible to accurately align the acoustic centers of the speakers only at one point in space, most often at the listening position. Naturally, the speakers' side radiation will no longer provide the necessary level of transient suppression for an FIR filter. Solutions to this problem lie in reducing room reflections, especially at the frequencies where the speakers operate together, and in shortening the FIR crossover to the minimum acceptable slope.
If very steep crossover slopes aren't necessary, the easiest solution is to switch to subtractive crossovers, which achieve a linear phase and eliminate long transients.
I've looked into that too. Imo, subtractive IIR crossovers are a heck of a lot of work compared to an easy-peasy linear-phase xover. And subtractive ends up having the same latency as lin-phase FIR (given both have the same xover order).
My vote is steer clear even for shallow slopes, unless you just don't have, or don't want to go with, FIR. Maybe I'm just lazy
Similar threads
- Locked
