REW and rePhase, tweak IIR with FIR or go all FIR?

[sfogg]

Within the last couple of weeks I made a bunch of changes to my system which includes a BSS Blu160. It does a bunch of things along with bass management (dual subs) and bi-amping and EQing my L/C/Rs. Last time I did this was *ages* ago using ETF and other crossovers. So I am also learning REW, Sound Architect and rePhase.

I am in the processes of tuning the speakers now and am just looking to get opinions on what direction to go?

Using traditional PEQ and NTM52 crossovers in the BSS I am making pretty good progress so far. But I just started playing around with rePhase and have FIR filters available in the BSS. I have never used FIR before testing it yesterday.

For those that have experience with both would you keep tweaking this system as is using IIR PEQ and crossovers and then try a final correction correct with FIR? Or start over and do crossovers and EQ directly in a single FIR filter?

I like the idea of being able to go really steep (my old setup used eliptics) but not sure if I have enough DSP in the box. I can likely free up a fair amount of DSP by removing some of what would be redundant processing objects. And I have a second Blu160 that I could put in to process just the L/C/R.

Right now everything is using NTM52 crossovers at 80hz to the dual subs and at 500hz between the bass horn and high frequency horn. I believe I am using 2 PEQs on each sub, and 2 on the bass horn and just a shelf filter on the high frequency CD horn.

Here is where I am at so far....

I *think* this is mostly time alignment needing to be tweaked a little more between the bass horn and the 2 subs.

Thanks for any thoughts or suggestions.

EDIT: While testing I think I hit the woofer in the bass horn with something it didn't like. I am getting a touch of a VC rub occasionally. It is about 40 years old so can't be too upset about that. I just ordered 3 new woofers to swap into the bass horns. They are supposed to have a bit more energy near the upper crossover point which would be helpful.

 

[Keith_W]

How many FIR taps are available with your BSS Blu 160? Unless you have more than 16k taps per channel, IMO don't bother with FIR full range. Otherwise, you could consider using IIR for minimum-phase bass correction and use very broad FIR filters for the top.

 

[sfogg]

Thanks, good question. This is the full layout in my one BSS160. This was at 77% DSP utilization before swapping some of it to FIR



This would be the six filters for each driver in L/C/R. The subwoofer high pass (80hz) happens before these and subwoofer correction would likely be IIR.


If I set the coefficient for each of the 6 to 2048 it still fits in my available DSP.


If I go higher than that the DSP resources jump up really quickly. I added an empty Blu160 and it looks like 6700 is as high as it will let you go in any one processing block. 4 FIR filters at 6700 maxed out the DSP. That should be running at 48k.



Thanks.

 

[Keith_W]

I suspect that is the best you can do with your DSP unit. You don't have enough FIR taps for bass so it will have to be IIR only.

 

[sfogg]

Thanks, is there a formula to know to better understand how the taps relate to frequency?

What about using FIR for the crossovers at 500hz? In rePhase with 2048 taps the high and low pass curves looks like it is matching the target well for pretty much any slope besides brickwall. Or is it just not worth the hassle?

Thanks again.

 

[Keith_W]

Have a look at this thread.

A "tap" is a time domain entity, and a "bin" is a frequency domain entity. You calculate your bin size with a simple formula: Bin Size = Sample Rate / Taps

So for e.g. you use a sample rate of 48kHz and have 2048 taps, your bin size is (48000/2048) = 23.4Hz. As you can imagine, the more taps you have, the better. The lower your sample rate, the better. Within limits, of course! You don't want so many taps that your latency goes through the roof, and you don't want a sample rate so low that Nyquist is below the audible spectrum.

Re: FIR vs. IIR for XO's. A minimum-phase slope rotates phase by 90deg for every order, so it makes XO design a bit more complicated. Some combinations don't sum to flat, some will require you to invert the polarity of one of the drivers, etc. Also don't forget that the mechanical bandpass of your driver is also minimum-phase, so the final result is a convolution of the minphase XO and the minphase slope of the driver. So if you design a 4th order XO slope and combine it with a 1st order driver roll-off, the final result is a 5th order slope, which may not sum to flat.

A linear-phase slope has no such problem. ALL symmetrical slopes of ANY Q will sum to flat because there is no phase rotation. You can also flatten the minimum-phase behaviour of the driver so the final result is truly linear phase and will definitely sum to flat. The problem is whether you have enough taps! Also bear in mind that if you choose a steeper slope or do a lot of phase correction, you can get a lot of pre-ringing. Without doing the simulation I can't tell you if what you are proposing is a good idea. I work with 64k taps so I have never encountered a problem, I just do what I want and check for pre-ringing. With only 2048 taps, I don't know. Try it and see.

 

[sfogg]

Thanks for the help, I will check out that thread and proceed cautiously. I definitely want to watch latency as these three channels are part of 16 others that would be time aligned. If I push too much delay on all of them I might have some lip sync issues.

 

[sfogg]

Just as an update. I was able to fit FIR filters on the woofers and tweeters of my LCRs as well as for my L and R subs all in the Blu160. 2048 taps for each.

The subwoofer low pass is fourth order at 80hz. Can't do any correction with FIR as the taps limited it. But since room should be minimum phase issues I will correct the same way.

Woofer high pass is fourth order at 80hz (this is where the taps limited me from going steeper as the match was a mess)
Woofer low pass at 500hz is eighth order and I can do EQ in this range with FIR.
Tweeter high pass is 500hz, eighth order, can EQ with FIR though don't really need any.

Think I can tweak this slightly more with time alignment.

 

[gnarly]

For those that have experience with both would you keep tweaking this system as is using IIR PEQ and crossovers and then try a final correction correct with FIR? Or start over and do crossovers and EQ directly in a single FIR filter?
Good stuff sfogg, I'm also into experimenting with FIR and IIR implementations .....on multi-way DIYs.

Like Keith says, a good strategy is to use IIR PEQs for sub/low frequency work. And FIR past that. Sytsem high pass, sub high pass, is best IIR always I think.
I try to make the xover between sub and main speaker complementary linear phase. If you do all the sub's response smoothing with IIR, and perhaps the woofer's too, you can use FIR just for the lin phase xover. FIR crossovers alole with no PEQs embedded can get by with fewer taps than expected. I think because xovers have pretty smooth transition everywhere. 2048 taps @48kHz will work, with optimization and window choice in rephase. 4k taps will make it easy id the BSS allows.

For a long time I simply used same size FIR file on every drivers outputs. Mainly for ease and an automatic target matching FIR generator like from Eclipse audio makes is so dang easy. I use qsys core 510i which allow 16k taps per channel @ 48kHz (for 15ch on 5-way LCR setup). And I used 16 k taps on every channel simply to avoid having to put in fixed delays that would be needed for different FIR size.

BUT !! In taking measurement and tuning classes, I learned the value of frequency dependent FIR file size that calls for fewer taps as frequency increases.....and how electrical filter impulses clean up the shorter a FIR file is. So lately, I might use 8k taps for the sub, 2k for the low, 1k for the mid, 512 taps for HF, and only 256 or 128 taps for VHF. Must say it really does sound a bit overall cleaner/clearer, and quieter in between notes.

The other major effort has been using IIR "preconditioning" on each driver section prior to apply FIR finishing/acoustic target matching. Clearly helps down low as already discussed. What's interesting and new to me, is for HF/VHF work...IIR does not seem to be the ticket compared to FIR, even when using very smooth low Q IIR filters.
Looking at the VHF driver's acoustic impulse response post filters, there is more gack, impulse oscillations, with IIR than with very short FIR filters like 256 or 128 taps. (on a 4kHz up driver)

Anyway, lots to learn and play with.

 

[sfogg]

Definitely lots to learn and play with. Thanks for the extra info. I could do 4k taps but I think I would run out. I'm running 8 channels at 2048 each. Not sure I can drop lower but I will see what the models show.


One thing I keep running into is sometimes the rePhase coefficents seems *way* off from what is predicted. Dangerously off.

I'm sure I'm doing something wrong but not sure what yet. For example, my high freq. horn runs from 500hz and up. If I model an eight order HP crossover in rephase at 500hz and load the coefficients into the BSS it works perfect and running a sweep I can clearly see the crossover just as predicted.

If I then leave the crossover settings the same in rePhase and also try to add some boost to the top end (basically a high shelf for the CD of the horn) and generate a new set of coefficients the predicted curve is a very close match to what I model. However, if I load those coefficients and then run a test (volume turned down for safety) from 400-20000 the crossover isn't there at all. It actually has massive boost at 400hz then drops down like a V to somewhere around 700hz and then has massive amounts of boost up to 20k.

Not sure if rePhase is messing up the coefficients or if the BSS doesn't like something about them?

I have had this sort of thing happen a few times, also going steeper has done it too. I'm in the habit of any new coefficients of test at a very low level and setting REW for the test frequencies to be a very safe range till I know they are working the way I want.

For my bass horn (80-500) I have a high pass, low pass and some EQ (dropping, not boosting) and those coefficients worked great.

 

[Keith_W]

One thing I keep running into is sometimes the rePhase coefficents seems *way* off from what is predicted. Dangerously off.

It might be the nature of IIR. Read this article: Understanding IIR differences between brands.