TOPIC:

I'm afraid it's irrelevant at this point.
If your objective is to debunk or shine a light on the goofiness of the high-end audio industry you're way late to the party.  That ship set sail a long time ago.  People just laughed at people like Noel Lee when he emerged onto the scene.  This has become a strange industry with all sorts of intellectual dishonesty and self-deception.  Marketing is king and drives everything.

It is what it is.

Dave.

I’m not under any illusions of winning the war via forums posts. However, the world of DSP is delightfully short on magical audiophile assertions, and it’d be nice to keep it that way.

Hi guys,

I appreciate all the comments.  I am also an engineer (mechanical) so I enjoy such arguments/discussions. 

My key points are:
1) -10+0 sounds differently from 0-10 with Dirac OFF.  -10+10 sounds differently from +10-10 and 0+0.  Try it in a quiet night.  I'm pretty sure you can hear it.  
2) I agree with notman that "It’d also be nice if they came up with a device that’s capable of more than 2V RCA / 4V XLR output."  But SHD itself is already an excellent product.  I love it.
3) I don't think I "fixed the problem."  I will, though, recommend people try -10 at input instead of -10 at output. Digitally they are the same, but you may prefer one over the other. 

I did some web research and found other brand's Dirac devices are also clipping so it's not minidsp' fault.  For me I am ordering a DAC (RME ADI-2) that can handle much higher output voltage.  It's $$ but I have nothing better to do anyway. 

Maximum of 2 volts or 4 volts outputs shouldn't normally be an issue.....with normal sensitivity speakers and power amplifier gains in the 28db area.

The problem arrives because with some recordings (or streaming sources) the nominal level can be quite low.  So, unless you're going to fool around with Replaygain, or normalizing, or temporary gain increase, or etc, you can find yourself running out of volume control.
With the standard SHD unit, at least you have the option of balanced outputs which yields 6db more signal level, relatively.

You shouldn't have to resort to using an expensive outboard DAC.  I would use a simple analog line-amp scheme.  Much cheaper.

Dave.

I won’t make any attempts to fool myself, thank you.

dreite is right, it makes no sense to add a $1300 DAC to your equipment. The SHD holds its own against the RME ADI-2 in the ASR measurements. In fact, the SHD is ranked just a hair higher in SINAD (though admittedly performs a hair worse in multi-tone, but the point is it’s a dead heat). Why have two DACs like that?

Check out the Topping Pre90. It’s $600 and state of the art and capable of +16db gain on the BAL outputs and a whopping 16V output (ASR and Stereophile). Set your miniDSP to the strongest output signal it can do without clipping with Dirac applied, then get every last bit of lost gain and then some back in the analog domain with the Pre90. This also gives you a buffer to correct for naturally quiet sources, e.g. for me YouTube out of my Roku is lower than any other video service.

I would already have one myself except I would need two because I run a subwoofer and managing to keep the volume in sync across both would be a headache and I can’t justify the cost since I’ve figured out how to properly approach the SHD and Dirac. Also, it would annoy me to no end that miniDSP has still not delivered a fixed volume mode to lock the unit from adjustment for cases like this and home theater pass through (literally one of the longest outstanding requests they have for SHD). How hard could that be to implement with their desktop software? They tell us to just not touch the remote or front panel but in the real world we are forgetful or have kids or guests and things get bumped inadvertently and I don’t want to always have to check it’s where I left it.

edmundpan wrote:

I will, though, recommend people try -10 at input instead of -10 at output. Digitally they are the same, but you may prefer one over the other. 

And one more thing, stop saying “at input.” You’ve been shown the signal flow. You haven’t done jack squat to the input. You’re confusing yourself and potentially others. 

It should be obvious that it’s impossible to have control of the input. Say you’re sending 3V RMS to the RCA inputs of the SHD, you can’t magically un-saturate it. That logic goes for every audio device imaginable. 

The routing matrix is part of the same 32bit/96khz digital processing where everything else inside the SHD happens. It’s all far more akin to output than input, if you have to think of it that way. It’s just that miniDSP puts the matrix next to the input meters (and they really shouldn’t have inactive sliders visible there) so you think it’s related. Can’t blame you there really, bad UI design by them. Hopefully the new software console they’re rolling out now helps untangle this for people.

Even $600 for a Topping Pre90 is going way overboard.
There are numerous excellent products in the pro-audio world that could implement gain and/or balanced/unbalanced (or vice-versa) conversion on the SHD outputs.

Dave.

Thank you guys!  You are extremely helpful!

Hi,

this is very interesting. May I ask two questions?

How do you know that the input gain is applied in the routing matrix and not in the inputs section? I do not see such an info in the doc.

If you apply EQ filters, e.g. raising a frequency by +3db, isn’t it important in this case to lower the input (so before the filter processing is done) by -3db to avoid clipping?

Ok, and a third one: Do I understand it right, that dirac is handling the input gain itself automatically to avoid clipping if I raise a frequency in the target curve by some db?

Thanks
Gert

gbr wrote:

Hi,

this is very interesting. May I ask two questions?

How do you know that the input gain is applied in the routing matrix and not in the inputs section? I do not see such an info in the doc.

You mean, how do I know that the routing matrix, which has four controls and is used to prepare the four outputs for which of the two input signals they will pass, has it’s own set of gain controls that we can see and touch and use, and is clearly diagramed as such, is actually where the gain is applied? And why do I think gain is not applied in the input stage, where that makes no logical sense and isn’t diagrammed as such? I guess it’s because it makes perfect sense and I see zero reason miniDSP would lie about that?

Not only do I think “input gain” is sort of a made up term you’re using, you can clearly see in the diagram that nothing happens to the signal until it hits Dirac, if Dirac is applied. Is that so hard to accept?

If you apply EQ filters, e.g. raising a frequency by +3db, isn’t it important in this case to lower the input (so before the filter processing is done) by -3db to avoid clipping?

No. It doesn’t need to do anything like preparing the input signal that way. You’re imagining Dirac taking the input signal and sort of “softening it up” by turning everything down by -10db first and then rubbing its hands and getting to work.

That’s the whole beauty of a floating point digital environment. Your source material can come into the inputs at a hair’s width under 0db average across the frequency range — this is a lot of professionally produced mass-market content these days — and the SHD lets you pile on heaps of digital gain whether Dirac or PEQ or routing matrix controls or output controls. Floating point let’s it shove down everything relative to these boosts, and clipping within this digital domain is for all purposes impossible. Your limitation on clipping will only occur in the conversion to analog and the output voltage limitations. For us it’s 2V RCA / 4V BAL.

Ok, and a third one: Do I understand it right, that dirac is handling the input gain itself automatically to avoid clipping if I raise a frequency in the target curve by some db?

I hope this is answered above. There is no “input gain” and Dirac doesn’t need to do any such thing. Dirac analyzes your room and gives the processor a set of instructions to implement including gain boosts of up to +10db and the processor does as it’s told in the floating point digital domain, and so you’re left with a remapped frequency response curve and your only limitation on clipping is ultimately your analog output voltage, defined by a combo of source material, Dirac filters, routing matrix gain controls, crossover, PEQ, output gain controls, and master volume, in that order, per the diagram. 

This is what I’m talking about with moving around poker chips. You’re digitally adding 10db here, subtracting 10db there, implementing a crossover that’s subtracting immense gain values, then fine tweaking with PEQ or whatever, then tailoring all to your on-the-spot listening desires with your remote control, and it’s all happening in a completely nondestructive, impossible to clip, reality-does-not-apply environment where a ledger is being kept step by step and all that really matters are the final values that hit the DAC.

gbr wrote:

If you apply EQ filters, e.g. raising a frequency by +3db, isn’t it important in this case to lower the input (so before the filter processing is done) by -3db to avoid clipping?

 

I wrote a lot very quickly, so I already have a revision. That is sort of what is happening, and you weren’t as far off as I replied. Sorry to not give you more credit on that one. What you are describing is true if the given frequency being boosted is applied to a frequency that appears as 0db. But imagine your input is -15db at that frequency. The processor doesn’t need to go lowering anything to apply 3db of gain to a -15db signal. 

Basically, 0db is the reference point. If the DSP needs space to add 15db to a signal that’s somewhere north of -15db in strength, it shoves everything down relative to the rest, and then applies the gain. If it doesn’t need to, then it just adds it and carries on.

Eg, I have a turntable signal — 4mV cart into 38db gain phono stage — the signal comes in weak. I apply 12db gain across the board at my output controls to get it happily inline with my other sources (use Preset 1 for all digital, Preset 2 for turntable, because we can’t set input-defined gain controls). The DSP does not need to do any mental gymnastics to apply that gain to a signal that’s south of -12db across the frequency range. It just applies it. The same would go if I created a PEQ that just did +12 db to the bass or something but left the rest alone. The digital values assigned to the analog input when it goes from the ADC have the room to work with so it’s all very easy. 

@notman,

Thanks for taking the time to explain things. The wider community is reading this thread any trying to learn from it so please don't get annoyed when we ask daft questions! 

Reading your response I've learned something about 'floating point' and I think what you're saying is it allows such a large swing between quiet and loud that if the software detects you're going to hit a max or min limit it simply moves the whole range up or down to avoid it.

Now I went and Googled what the dynamic range of 32bit floating point and got 1528db as the answer. Presumably '0' sits in the middle of the range. Wow... How can a limit ever be hit that would require the software to shift the entire range up or down when there is so much range to begin with?!

I think I understand now why saying we should set the outputs at -10db to compensate for dirac's possible boosts is misleading as, as you say dirac may very well be cutting and a boost could come from somewhere else and/or even if dirac boosts, the content we're listening too and filters we've added may have 'cuts' greater than any boost dirac has applied. 

But I am left a little confused on how I can still clip. And I'm pretty sure I have clipped using the SHD. I've listened to music that sounded gritty until I reduced the outputs by 10db and the grittiness went away. I assumed this was clipping occurring although I'm confused how as presumably it would mean I was using the full 2v output which I couldn't have been because the input sensitivity of my amps is 0.6v and I wasn't playing things that loud. Perhaps it wasn't clipping but tye grit went away as I say, when I reduced the output by 10db.

You wrote that "your only limitation on clipping is ultimately your analog output voltage, defined by a combo of source material, Dirac filters, routing matrix gain controls, crossover, PEQ, output gain controls, and master volume, in that order" but I thought we're saying all those things occur in the digital domain and as such haven't we just said, try as hard as we might we can't clip there? So maybe you couod tell me if this assumption is correct and if not put me right.... Does the digital 0dB on the outputs tab equal 2v/4v analogue out? If so then although floating point guarantees no digital clipping, presumably it doesn't guarantee that the level stays below 0db and if the level doesn't stay below 0db dugitally we clip on the analogue output.

Thanks
Alex



 

Appreciate the kind words. I’m certainly no engineer, just a hobbyist who’s very enthusiastic about what DSP allows us to do and I’ve tried to learn. To be honest, we’re already solidly butting up against my suitability lead a discussion at all on this stuff, but I think have things straight enough to try. I just hope you find this remotely helpful and I’m not too far off the mark.

0db is always the very top in the digital domain, not the middle. There is no value higher than 0db, and 0db directly corresponds with the max output voltage on the device.

I’ll admit now I’ve played a little fast and loose on the master volume, and that’s the wild card here. Let’s try to sort this out.

Imagine this, you’re using REW and sending pink noise to the SHD via USB at “0db” as verified by the input meters in your SHD software. You do absolutely nothing to any of the gain inside the DSP and run the master volume at 0db. Simple enough, you should not be clipping the SHD output because the signal arrived unclipped at 0db and should be able to leave that way at 0db with no DSP hijinks involved.

In that example, with a standard consumer power amp being the next device in the chain, what you would very obviously hear is clipping through the speakers, but it’s because a consumer power amp’s input sensitivity is usually about 1.4 or 1.5V and in our current example the SHD output is operating above that, very near it’s 2V max, and saturating the amp inputs.

Now, so that we can rule out the amp inputs clipping, imagine a power amp with a much higher input sensitivity of 8V. Now the SHD is playing the same 0db pink noise via digital input. First we drop the SHD master volume to -10db, and then we add +10db PEQ peak filter somewhere with a lot of energy like 100hz. Again, simple enough to understand, we’re still good here. The SHD output isn’t clipping. We took -10db off the master volume but added +10db on PEQ and gets up back to zero. No problem.

But now we start to nudge up the master volume. We should very quickly have a problem, and it’s exactly what the whole “leave -10db for Dirac on your volume” is about. The value of that pink noise, first at exactly 100hz, becomes unrelentingly stuck at 0db/2V output, and the master volume is trying to force the analog stage to give more. And as you continue to drive up the master volume, now the range of maxed out frequencies is spread across 50hz to 150hz, and the analog output stage of your SHD is very, very unhappy and it all sounds absolutely terrible.

This is where I get a little unsure of myself. To me, what we’re describing is the analog output stage clipping, but now I’ve written this all down for this discussion, I could see someone saying, “but isn’t that digital clipping because the value was 0db before we over-applied the master volume, and now the master volume is trying to apply a digital value higher than 0db which it can’t, so it’s a digital clip.” I don’t have the chops to answer that definitively, but since 0db is 2V and 2V is 0db for us, it feels a bit like chicken and the egg.

In the real world, you have music, and maybe it arrives quiet hovering around -30db on your input meters, and all of your Dirac and PEQ and digital gain in the software doesn’t exceed +15db total, and maybe that gain isn’t applied in an area of the audioband with a lot of energy. In that case you’re could likely run the master volume all the way up to 0db and wish you had more output because you’re still too far under your amp input sensitivity voltage and/or your speaker sensitivity.

Also in the real world, you are listening to Billie Eilish via digital, and your input meters are absolutely pinned, and Dirac alone is trying to work magic on a huge, wide null in your room at 60hz with a huge, wide +10db boost, and the second you even attempt going up to -9.5db on the master volume you’ve hit the limits of full output and, your pick on how to define this, either the digital finally gives out or the analog stage does but it sounds awful.

That’s my best understanding. The best case scenario we get a real electrical/digital audio engineer around and they set the record straight and debunk some of what I wrote (I hope not too much of it!) and we all learn something.

@notman,
Makes sense. Easy to follow examples. Thanks. 

0db is not a level.....unless it's referenced to something.  Don't get confused by this.
0dbV = 1 volt RMS, 0dbV = 2.2dbu, etc, etc.  Otherwise it's a meaningless number regards actual signal level.

In the simplest case.....if you program and/or input enough level in the DSP portion to fill up the DAC input with all 1's, the output (of the DAC) will clip.
Normally, any analog stages following a DAC will be designed to have a higher voltage clipping point than the DAC output itself.  Thus, any clipping you experience will be "digital" clipping.

Dave.