MiniDSP - Page 48

I don't (yet) have a miniDSP, but I understand how their filters work.

It sounds to me as Don is referring to the fact that IIR (Infinite-Impulse-Response) filters have a non-linear phase response. In other words, different frequencies are delayed by different amounts of time. FIR (Finite-Impulse-Response) filters, on the other hand, can have a linear phase response: a constant delay for all frequencies. This constant delay, called "group delay", is equal to the time between each sample, multiplied by half the size of the filter. This means that, if you use the same size FIR filter for each channel, everything comes out at the same relative phase as it went in, although delayed by the group-delay.

The biquad filters used by the miniDSP are a specific form of IIR filter, and therefore they will suffer from unequal delays for different frequencies. But the phase-shift is small, and some say its of no consequence (I cannot speak to that issue - I just don't know). It can also be mitigated somewhat, and a Linkwitz–Riley crossover is an excellent example of how to repair the phase delay.

Note that the time-domain adjustments in Audyssey or the miniDSP might not be useful unless they can be made to effect the delay with the inverse relationship to frequency as the FIR filter's effect.
Edited by MarkHotchkiss - 2/4/13 at 1:11pm

Thanks for your time and help. I have some learning to do but this is helping me figure out what I need to figure out.

One note for those that were discussing over in the other thread, the auto EQ REW can send to a minidsp uses the biquad filters so as stated above, there is an impact but may be minimal enough to be inaudible.

Can anyone answer either of the first two questions? Sorry to derail the interesting and much more technical discussion going on here:D

Came straight to my door either ups or FedEx. Don't remember which. Signature required too Im pretty certain.

I bought mine about 8 months ago (and here I am just now looking at actually using it, long boring sub story)

1. Mine was shipped straight to the door
2. Not sure who actually delivered it but it had a hong kong shipping sheet "EMS something" on it. Mostly english but some mandarin.
3.The price I paid directly to minidsp is all I paid. I hope the dealer is taking care of that sort of thing since they have a sales website pointed at the US. Potentially they are sneaking around it which I don't really think is kewl. May be worth asking in the forums on their website. If you find out I'd be interested to know.


HeHe that technical talk is hurting my brain and I'm the one asking. Must read more.

Is the miniDsp a decent devise for use as an active crossover? What all is involved in crossing over actively? (note that I know about the hardware needs, amps ect.., just trying to figure out what and how to make the settings in the miniDsp)?

Thanks for the shipping info guys! That's just what I needed to know.

After asking my boring shipping question above, let me ask something a little more technical and on point for the thread.

If you're using the miniDSP for EQing a sub, do you think there will be any audible degradation to the sound by allowing a receiver to do a digital to analog conversion, and then sending that signal over to the miniDSP for another round of conversions (back to digital for processing and then back to analog for output to the amp. I think I have all that right.

Given my current setup, I don't see a way around this, but wonder if it's worth looking into alternatives to avoid the extra conversions. My understanding is that the conversions are a lossy process, so this is why I'm curious. I'm less concerned for HT, but would like to use my sub (LMS-Ultra) for 2.1 listening as well.

it is made to be a active xo i would also buy a mic from them to go with it so u can measure the speakers and apply a good xo without any phase or any other problems that come with blindly going at it.

Hi all,

I noticed there is a version of the miniDSP that does FIR filters, though I did have the IIR in mind...

Others have nicely addressed my Q&A already. I'll just throw out two quick thoughts:

1. Correcting in frequency domain may not help time alignment, and may hurt as has been stated due to the additional delay and nonlinear phase of the filters. I suspect this is minor.

2. If I want to time-align the wavefronts from say the sub and main speakers I may need a frequency-dependent phase shift, and/or a true time delay. That could be accomplished by an all-pass filter after the crossover in the DSP (or where ever). I have not looked deeply into Audyssey, but MCACC does do phase adjust, and I would like something similar with the miniDSP. Just not sure it's in the supplied SW.

Note time alignment of the speaker outputs is a seperate (though perhaps related) problem than crossover phase alignment, which a L-R solves nicely.

The simplest example is a sub located away from the mains, thus the simple min phase filters or an L-R crossover does not work; you need to delay the signal to the sub or mains.

The ugly example is multiple drivers physically offset, or perhaps a sub and multiway speaker where the main speaker is not itself time-aligned due to speaker (driver) physical placement, crossover phase shifts, or both. Now you essentially need to adjust the phase/group delay across frequency to time-align each frequency point. Could get nasty, or at least DSP-intensive.

Except to make pretty pictures of step/impulse response (which my Maggies do well, thank-you-very-much), the whole issue may be moot in the real world of music and movies. But, the design engineer in me wonders about such things... Too many years dealing with systems where time response was critical (radar, lidar, sonar, ultrasound, etc.)

Hi Don,
It's not very hard to compensate for physical offset, as that compensation can be done with a fixed delay. A fixed delay is practically free from a DSP processing perspective, but in terms of memory resources, it has a moderate cost.

You simply hold your PCM output samples in a ring-buffer before sending them to the DAC (or other output device). With a 48kHz sample-rate, a 48 sample buffer will delay for 1 millisecond, which will account for approximately 13.5 inches of speaker displacement. You adjust the delay by adjusting the size of the buffer. Most (maybe all?) DSPs have direct hardware support for ring-buffers, which is why it takes minimal DSP processing.

Notice that the amount of memory required goes up proportionately with the amount of delay, the number of output channels and the sample-rate. That is why many DSP applications forgo the higher sample-rates.

How does a person time align their front LCR's and two subs using the miniDsp? If a person is measuring his fronts and duel subs using a UMM-6 through REW, which measurements in REW are used to time align and how does one go about properly applying these to the miniDsp in both an active and passive crossover situation?

There are two methods I am aware of, although I'm trying to understand how to interpret phase better which should also be a valuable piece of information. I think the first one is easier and the second one is better. Note the general approach is to Time align the furthest sub with the mains then introduce delay (or tweak phase) on the closer subs to be aligned with the furthest sub+mains. Note not all phase knobs are created equal, if it is not implemented as a true group delay it adjusts the timing differently at different frequencies, but then the best you can do with it is focus on that all important crossover range.

Sine wave at crossover point) Once you're happy with your levels (gain matched/level matched and calibrated level to mains) play a sine wave at the crossover point and tweak this distances / phase until you find the setting combination which produces the highest SPL (reading SPL of sine wave at crossover frequency). Start with furthest sub + mains, then add on next closest sub. This can be done with only an SPL meter, is better than nothing, and should be eliminating cancellation due to speaker interaction.

Frequency Response Inspection) Again get your levels right, then start with the furthest sub. The idea here is to run frequency sweeps around the crossover range (or wider sweeps and zoom in to inspect the crossover range). It's really not just the crossover frequency we care about but that entire transition range so aim for the smoothest transition possible. We prolly care about 20-30Hz around the crossover freq. Once your distance/delay/phase is set to produce the smoothest FR in the transition zone, add on the next closest sub, one by one and repeat the sweeps inspecting that range and tweaking.


Seems like understanding the phase plots would really help either of the above approaches out. In the end it is trial and error hopefully with a bit of an educated approach and examining REW plots to verify which is best.
Edited by dstew100 - 2/5/13 at 5:43am

Yes, of course, but does the miniDSP SW provide that? I am not trying to be a smart-****, just not sure what their SW includes and not sure the specs (memory size) of the on-board DSP. And yes for the money I should just get one and play with it, but I'm cheap and lazy... I am guessing it does, that's certainly one of the most basic criteria and easiest for them to implement.

Once you've accounted for physical distance, aligning the phase at the crossover point should suffice for most of us, I would think. More advanced phase/time alignment over frequency to account for other effects is interesting but probably more an academic exercise, at least for audio speakers.

Thanks Mark! - Don

p.s. Has anybody tried the new FIR filter box?

Most of the information you're looking for can be found in the spec sheets for the software plugins, the hardware spec sheet and the hardware manual.

What part of "lazy" did you not understand? OK, RTFM it is!

* The spec sheet for the 2 way plus sub crossover file says 0 to 7.5 ms per channel in 0.02 ms increments. I also notice it includes Butterworth, L-R, and Bessel; the latter is interesting. Lower rolloff for a given order, but linear phase (constant group delay).

I found this a little confusing:
Master output gain: Analog potentiometer control master output digital gain fader from –80 to 0dB. Disabled if no pot connected.

So, is it a digitally-controlled analog pot at the output, or a pot that is controlling the digital level into the DAC?

Old fart admission: I clicked on the PDF icon in the past and just got the list of plug-ins; I missed the datasheet link at the bottom of the table.

In section 2.4.4 of the hardware manual, they show the pot hooked up between 3.3Vdc and ground. It looks like the DC voltage from wiper to ground gets A/D converted and the resulting digital word used to control the gain of all channels simultaneously.

I haven't ordered mine yet, and I wasn't going to use that option, but now that I think about it, it might be useful. I'll be using mine for a multi-sub arrangement with 4 subs, so it looks like the mono four-way advanced plug-in is the one to get.

Does it drive a digitally-controlled potentiometer (or similar, essentially digital control of an analog parameter), or is it used to scale the digital words going into the DAC? In the real world probably does not matter, I am just curious. I should see what DAC they are using, that might provide a clue.

I don't think there's any analog gain control in it at all, so my guess is it's the latter.

I've been looking at it, "And yes for the money I should just get one and play with it, but I'm cheap and lazy..." also.

Only one product, the OpenDRC, supports the FIR filters. It is only available as a 2x2, analog or digital version with one plug-in. The plug-in works like this:

IN -> Input select -> Gain & Mute -> FIR filter -> LPF+HPF IIR filters -> 6-band IIR PEQ -> Gain / Polarity / Delay -> Master Volume -> Compressor / Limiter -> OUT
Yes. The "Delay" block allows for a delay up to 1 second (1126 feet), in 21 microsecond increments (which is .28 inch increments).

Hi RB, I suspect the same, based on the block diagram.

Thanks guys. Some chips have an on-chip analog gain block controlled by a digital word (so the analog signal is stepped in volume but the actual signal path remains analog, an MDAC sort-of) or analog voltage level (e.g. analog control signal varies the gain of an analog multiplier), so I was not sure what the miniDSP was doing. Others simply throw away bits to reduce the volume in a purely-digital system. Pros and cons could be debated endlessly.

The IIR-only models like the 2x4 have only a 7.5 msec max delay per channel (from the plug-in docs), which is shorter than I'd like to see.

Edit: Oops, the 2x8 has a max delay of 9 msec, and for the 2x4 it is 7.5 msec max.

IIRC some have 15 ms

u also need a way to generate the fir filter it doesnt come with the software to do it.

Hi CookieAttk, Ah, good catch. I did not notice that.

That would not be a problem for me, as I use MathCad to generate my filter coefficients. But it could certainly be a hindrance for others.

I like generating FIR coefficents, while I find calculating IIR coefficients to be tedious. One of the reasons I prefer FIR filters to IIR filters is that it is much easier to design the filter - you just need to "draw" the frequency response you want, then take the fourier transform, and then select and add the "window" function that suits the application. Besides Mathcad, Excel can calculate all this as well. There is a lot more involved with IIR filters, which may be why they provide IIR software, but not FIR software. Or maybe they just haven't gotten around to writing it yet.

I am having troubles with the biquads automatically calculated by REW. I input them into the minidsp and on all except the first and second filter result in a message saying that the biquad is not formatted correctly. Anybody alse have this problem?

I've never seen that. Although I've only ever attempted to eq the same setup, I did at some point do quite dramatic things to what REW had done, to get a sense for how predictable things are, and there was never any sort of complaints.

These are the biquads as generated by REW.
biquad1,
b0=1.000744204966523,
b1=-1.9984866417835156,
b2=0.997760299915597,
a1=1.9984866417835156,
a2=-0.99850450488212,
biquad2,
b0=0.9997439483092558,
b1=-1.999143157101479,
b2=0.9994052301550241,
a1=1.999143157101479,
a2=-0.9991491784642799,
biquad3,
b0=0.9996331436324764,
b1=-1.998512809008495,
b2=0.9989126634976113,
a1=1.998512809008495,
a2=-0.9985458071300877,
biquad4,
b0=0.9998139353882447,
b1=-1.9990697515212206,
b2=0.9992657223347333,
a1=1.9990697515212206,
a2=-0.9990796577229778,
biquad5,
b0=0.9995480188479403,
b1=-1.9982529907727384,
b2=0.9987869551076304,
a1=1.9982529907727384,
a2=-0.9983349739555707,
biquad6,
b0=1.0003641116189996,
b1=-1.9979589531660644,
b2=0.9977214473649936,
a1=1.9979589531660644,
a2=-0.9980855589839932

I've seen that, but only when I have not properly pasted in the filters. You can only put one biquad into each PEQ band. And in each PEQ band, the biquad should start with "biquad1". If you don't change the biquad# number to a 1, it will fail.

For instance, this does NOT work:


And this DOES work:

But if you have a text file with all of those listed, biquad1, biquad2, etc - you can use the "Import REW File" button, and it will do all the data entry work for you. After importing the REW text file, it will put each set of filters in a different EQ band, and they will all begin with biquad1. I just created a text file with your numbers to test, and it imported properly.