Review and Measurements of Emotiva XMC-1 Gen 2 Pre/Pro

[DonH56]

Yes, you are correct. For Analog input, we do have a "Reference Stereo" mode that will literally take and Analog input, feed it through the volume controls and send it out. As you stated though, it is only for Analog.

Lonnie

Whew! Blind pigs and acorns, gotta' get one right sometimes. It sounded like you were speaking of the digital signal path so I wanted to clarify there is a "straight wire" analog path for those who want it.

Appreciate you stopping in here, Lonnie, as well as all of your contributions to the Emotiva Lounge.

 

[Lonnie]

Except that many AVRs purportedly operate in 2.x and 5.x mode for playing music. Many AVR manufacturers use 2 channel specs in their power ratings for this (and other) purposes. Additionally, AVRs are configured for streaming music.

So while I think there is a weak case to be made for exempting something like sound bars from testing at an audiophile standard, that case evaporates when we look at how AVRs are configured and marketed.

I think it is pretty easy to test an AVP vs. AVP. Generating multi-channel test patterns is not that hard and you can easily see, the broadband frequency response, noise floor, dynamic headroom, THD and such. Now how AVRs are rated is a bit of a mess in my opinion. Most only state a power rating for one or two channels driven no matter how many channels it has. I guess this is fair if all the other manufacturers do it, but there are no guidelines or set standards for this. Personally I like to rate my amps "All channels driven", but that is just me.

Lonnie

 

[GrimSurfer]

I think it is pretty easy to test an AVP vs. AVP. Generating multi-channel test patterns is not that hard and you can easily see, the broadband frequency response, noise floor, dynamic headroom, THD and such. Now how AVRs are rated is a bit of a mess in my opinion. Most only state a power rating for one or two channels driven no matter how many channels it has. I guess this is fair if all the other manufacturers do it, but there are no guidelines or set standards for this. Personally I like to rate my amps "All channels driven", but that is just me.

We all know how the marketing literature is written. Its intent is to either land the sale of justify it, sometimes against advice provided by the engineering dept.

The consumer electronics industry seems pretty comfortable interchanging AVRs and stereos in their advertising and at the point of purchase, so I'm disturbed when subsequent arguments based on "exception" are made.

Now to be clear, I have posted elsewhere that it is absurd for anyone to believe that a multi channel device like an AVR under $3k will have any hope of delivering exceptional measured fidelity. Once we crest that price point, however, scaling is possible. But until the industry has that conversation with consumers, it will be up to folks like @amirm to point out just how poorly (or well) devices like AVRs deliver on promises that are both stated and implied.

 

[amirm]

With all due respect, I'm sorry to say that I do not know of any processors whose Pure Direct mode actually bypasses the DSP. What Pure Direct does, at least for the various manufactures that I know, is simply this. The signal still has to go through the DSP to be decoded and rendered. Depending on the manufacturer, bass manager is sometimes dis-engaged and sometimes not. EQs are set to 0, and depending on the manufacturer, room EQ will or will not be dis-engaged. Time delays will still be retained through.

Pure Direct mode for most manufacturers (Including us) will retain bass management and time delays and turn off EQs, Room Correction and things like that.

I set all speakers to large in my testing. And of course all EQ/DRC is turned off. Delay is set to zero. I realize that the PCM signals still go through the pipeline. But I like to see one where the internal resolution is high enough to not matter. I hear you on gain and that can be an issue but companies like Benchmark deal with that in their 2-channel products.

I hear you on issues of getting this pipeline right and we did that in Windows when we re-wrote that stack. But alongside also provided a pure, exclusive mode WASAPI interface for when no processing is needed. It shouldn't be hard to provide such a bypass path.

 

[RichB]

In my opinion, comparing any AVR/ AVP to a simple DAC is not a good comparison. AVR/AVPs cannot and will never perform at that level and I will tell you why. For a DAC the SPDIF signal comes in and goes straight to the DAC. After the DAC it goes through an I/V converter and out. Pretty straightforward, no post processing going on so nothing to degrade the performance. Now lets look at an AVP for comparison. Bitstream comes in and goes to the DSP. The DSP decodes and renders the various channels. The output of that then goes through a bass manager where crossovers are applied, the bass is summed and redirected. The dedicated LFE channel is also put into the mix where it gets summed as well. After that it all goes to EQs, Room correction, time delays and finally sent to the DACs. Don't be fooled into thinking that all this happening in the digital realm is any different from what would happen in the analog realm. DSP algorithms are just mathematical expressions of an analog circuit and since a DSP cannot operate above 0dbfs a lot has to be done to prevent it from clipping. You can't sum 16 channels of bass without lowering the level down quite a bit, which means you have to lower everything to match. Add in EQs, you can't boost a signal in a DSP, so what you are doing is actually cutting everything around the center frequency to allow the chosen frequency to appear boosted. Room correction works the same way as an EQ. I feel confident that most here are aware that when you lower a signal in the digital realm, you loose resolution. EQs and room correction add phase issues as does time delays. So all this summing, crossovers, EQs and delays mean a lot of signal manipulation before it ever ends up at the DACs. If you were to do everything being done in a DSP in the analog realm, you might be surprised at how bad your performance actually would be.

Bottom line, because of all the signal manipulation going on in an AVP/AVR, it will never equal that of a simple DAC.

Lonnie

Is there always a fixed S/N hit when processing is engaged or does it vary based on the amount of processing engaged?

- Rich

 

[RichB]

I set all speakers to large in my testing. And of course all EQ/DRC is turned off. Delay is set to zero. I realize that the PCM signals still go through the pipeline. But I like to see one where the internal resolution is high enough to not matter. I hear you on gain and that can be an issue but companies like Benchmark deal with that in their 2-channel products.

I hear you on issues of getting this pipeline right and we did that in Windows when we re-wrote that stack. But alongside also provided a pure, exclusive mode WASAPI interface for when no processing is needed. It shouldn't be hard to provide such a bypass path.

"Reference Multi-channel".

- Rich

 

[Blumlein 88]

In my opinion, comparing any AVR/ AVP to a simple DAC is not a good comparison. AVR/AVPs cannot and will never perform at that level and I will tell you why. For a DAC the SPDIF signal comes in and goes straight to the DAC. After the DAC it goes through an I/V converter and out. Pretty straightforward, no post processing going on so nothing to degrade the performance. Now lets look at an AVP for comparison. Bitstream comes in and goes to the DSP. The DSP decodes and renders the various channels. The output of that then goes through a bass manager where crossovers are applied, the bass is summed and redirected. The dedicated LFE channel is also put into the mix where it gets summed as well. After that it all goes to EQs, Room correction, time delays and finally sent to the DACs. Don't be fooled into thinking that all this happening in the digital realm is any different from what would happen in the analog realm. DSP algorithms are just mathematical expressions of an analog circuit and since a DSP cannot operate above 0dbfs a lot has to be done to prevent it from clipping. You can't sum 16 channels of bass without lowering the level down quite a bit, which means you have to lower everything to match. Add in EQs, you can't boost a signal in a DSP, so what you are doing is actually cutting everything around the center frequency to allow the chosen frequency to appear boosted. Room correction works the same way as an EQ. I feel confident that most here are aware that when you lower a signal in the digital realm, you loose resolution. EQs and room correction add phase issues as does time delays. So all this summing, crossovers, EQs and delays mean a lot of signal manipulation before it ever ends up at the DACs. If you were to do everything being done in a DSP in the analog realm, you might be surprised at how bad your performance actually would be.

Bottom line, because of all the signal manipulation going on in an AVP/AVR, it will never equal that of a simple DAC.

Lonnie

@Lonnie It seems to me if you did your summing and DSP in 32 bit float (or even better 64 bit float) there is no reason you couldn't get results like or very close to a simple high quality DAC. You don't lose resolution doing summing this way like you do in digital using 24 bit fixed. Most DAWs can do such things without it becoming a bottleneck of noise on the end result. It might require better chips and processors to do it real time, but I see no reason it couldn't be done at least in premium products.

FWIW, a friend has the XMC-1, and I usually set things up for him. I think it an excellent unit and combined with Dirac it really shines in a high quality system.

 

[Lonnie]

The DSPs are 32 bit but just like in the analog realm, there is a price to pay for every layer of processing. We spent more then a year just optimizing the system structure to minimize the negative effects and I believe we have done a great job of it. If you listen to one of our new processors, I believe you will be blown away.

 

[GrimSurfer]

The DSPs are 32 bit but just like in the analog realm, there is a price to pay for every layer of processing. We spent more then a year just optimizing the system structure to minimize the negative effects and I believe we have done a great job of it. If you listen to one of our new processors, I believe you will be blown away.

Cool. Let's see how they measure too.

Maybe you could send a unit to @amirm for independent testing?

 

[Blumlein 88]

The DSPs are 32 bit but just like in the analog realm, there is a price to pay for every layer of processing. We spent more then a year just optimizing the system structure to minimize the negative effects and I believe we have done a great job of it. If you listen to one of our new processors, I believe you will be blown away.

Is it float or 32 bit fixed? Float should have very little cost to the results that would reduce straight through measurements.

Also what was used in the XMC 1?

 

[DonH56]

Note 32-bit float has a 24-bit mantissa (assuming IEEE format) so when combining signals that may limit the "practical" dynamic range. Last time I tackled DSP stuff was years ago but it was not unusual to have +/-20 dB swings through the filter stages due to summing and the usual filter peak/null mess. That means you may be throwing away 3-7 bits or so to keep enough headroom through the DSP stages to ensure nothing clips. That is without considering the source mix, but is another reason 24-bit chains are common in the DAW world. Back in grad school much of the DSP filter theory coursework I took focused on how to optimize the dynamic range through cascaded filter stages.

I used to think most audiophiles would faint if they knew just what the recording signal path looked like. Being older now I have discussed it enough to know that, to them, it is not the bazillions of transistors and bits processed before their source file that matters, just how many from there to their ears. Same argument applies to things like power cables (miles of power lines to the house and hundreds of feet of wire inside do not matter, only the last three feet from wall to component).

 

[RichB]

Note 32-bit float has a 24-bit mantissa (assuming IEEE format) so when combining signals that may limit the "practical" dynamic range. Last time I tackled DSP stuff was years ago but it was not unusual to have +/-20 dB swings through the filter stages due to summing and the usual filter peak/null mess. That means you may be throwing away 3-7 bits or so to keep enough headroom through the DSP stages to ensure nothing clips. That is without considering the source mix, but is another reason 24-bit chains are common in the DAW world. Back in grad school much of the DSP filter theory coursework I took focused on how to optimize the dynamic range through cascaded filter stages.

I used to think most audiophiles would faint if they knew just what the recording signal path looked like. Being older now I have discussed it enough to know that, to them, it is not the bazillions of transistors and bits processed before their source file that matters, just how many from there to their ears. Same argument applies to things like power cables (miles of power lines to the house and hundreds of feet of wire inside do not matter, only the last three feet from wall to component).

A laudable goal is for minimal impact through the DSPs (all flat and no processing). Audible transparency is the goal but measurements help with the assessment.

Eschewing the term "audiophile", I top out at the vapors

- Rich

 

[Blumlein 88]

Note 32-bit float has a 24-bit mantissa (assuming IEEE format) so when combining signals that may limit the "practical" dynamic range. Last time I tackled DSP stuff was years ago but it was not unusual to have +/-20 dB swings through the filter stages due to summing and the usual filter peak/null mess. That means you may be throwing away 3-7 bits or so to keep enough headroom through the DSP stages to ensure nothing clips. That is without considering the source mix, but is another reason 24-bit chains are common in the DAW world. Back in grad school much of the DSP filter theory coursework I took focused on how to optimize the dynamic range through cascaded filter stages.

I used to think most audiophiles would faint if they knew just what the recording signal path looked like. Being older now I have discussed it enough to know that, to them, it is not the bazillions of transistors and bits processed before their source file that matters, just how many from there to their ears. Same argument applies to things like power cables (miles of power lines to the house and hundreds of feet of wire inside do not matter, only the last three feet from wall to component).

https://www.sounddevices.com/32-bit-float-files-explained/

Lower half of this page is the sort of thing I had in mind. Effective dynamic range of 1528 db. So you can do plenty of processing and scale everything so there is no problem with clipping or adding noise to the noise floor.

Of course adding gain to an input with noise will raise the noise, and I suppose the 24 bit fixed output of the DAC will need to leave room for portions of the processing which have gain. It seemed to me that if DSP is 32 bit float and we are running a straight through signal we could get better results than are being seen. There probably are good reasons it isn't being done that way. Like a straight thru volume level being set, and then you add a boost to your subwoofer of 6 db, and while everything is adjusted you'd get that 6 db boost by lowering the volume of everything else. This would make most consumers scratch their heads. This would be necessary even if the processing was in float and not impacting the results because of the bottleneck of 24 bit fixed DACs.

 

[DonH56]

https://www.sounddevices.com/32-bit-float-files-explained/

Lower half of this page is the sort of thing I had in mind. Effective dynamic range of 1528 db. So you can do plenty of processing and scale everything so there is no problem with clipping or adding noise to the noise floor.

Of course adding gain to an input with noise will raise the noise, and I suppose the 24 bit fixed output of the DAC will need to leave room for portions of the processing which have gain. It seemed to me that if DSP is 32 bit float and we are running a straight through signal we could get better results than are being seen. There probably are good reasons it isn't being done that way. Like a straight thru volume level being set, and then you add a boost to your subwoofer of 6 db, and while everything is adjusted you'd get that 6 db boost by lowering the volume of everything else. This would make most consumers scratch their heads. This would be necessary even if the processing was in float and not impacting the results because of the bottleneck of 24 bit fixed DACs.

The problem is when you have more than one signal at different amplitudes. The maximum range in that situation is set by the mantissa. That is a problem I have often dealt with in the RF world and seen it in the studio when mixing instruments and such (I have limited and old studio experience, however). That is why I used "practical" -- in the real world I am not sure (and I mean that literally -- I do not really know) how much floating point processing gets you for audio signals where the dynamic range tends to be <100 dB (and often much less) in the final mix. For signal-processing stuff I have helped design in the past, dynamic ranges of 120 to 160+ dB were pretty typical, low-resolution converters were used for speed, and floating-point processing was beneficial since the analog front end could help isolate signals into narrow bands. At the end of the day, the DAC will set the dynamic range, so maybe floating-point along the way helps? Again audio signal processing is not my day job so I am not sure.

 

[Blumlein 88]

I just did this both in Audacity and a DAW. Put more than one signal in as 32 float. Knocked it all down 500 db. Did various processing steps including summing it all when done. Brought it back up just short of clipping. No added noise or other artifacts above a 24 bit noise floor. So such is at least possible whether simple or working in a way a consumer would want an AVP to function. The noise and distortion of the DAC I use to output this is the limiting factor not any of the processing or summing. Any noise in the summed signals will sum as an increased noise floor, but a straight through unprocessed signal thru such a setup need not hamstring basic performance.

 

[DonH56]

Again, it is a two- or multitone problem when one is full-scale and one at or below the range of the mantissa, ~2^-24. No "analog" noise is added in this case. It is a big problem in some systems but I am not sure it is a practical problem in audio as the signals typically do not have that much dynamic range. The quantization-based SNR of a 24-bit system is about 146 dB so I cannot really imagine this being an issue in the real world. For your test, the equivalent would be one signal at 0 dB and the other at -500 dB. Sum them using floating-point math and the smaller one will likely disappear.

Completely agree the DAC, and other components and our own ears, will likely be the dynamic range limitations in a real (audio) system. There are practical and theoretical limits and I cannot really see 32-bit integer arithmetic losing much if anything over 32-bit floating point for an audio system.

 

[Blumlein 88]

Again, it is a two- or multitone problem when one is full-scale and one at or below the range of the mantissa, ~2^-24. No "analog" noise is added in this case. It is a big problem in some systems but I am not sure it is a practical problem in audio as the signals typically do not have that much dynamic range. The quantization-based SNR of a 24-bit system is about 146 dB so I cannot really imagine this being an issue in the real world. For your test, the equivalent would be one signal at 0 dB and the other at -500 dB. Sum them using floating-point math and the smaller one will likely disappear.

Completely agree the DAC, and other components and our own ears, will likely be the dynamic range limitations in a real (audio) system. There are practical and theoretical limits and I cannot really see 32-bit integer arithmetic losing much if anything over 32-bit floating point for an audio system.

I think we are in agreement. The various multi-tones I used covered a range of 100 db. I dropped them down to EQ, and then sum before bringing them back up. I of course don't care about any signals 500 db apart in size. My point is that such processing done in 32 float won't prevent an AVP from achieving just to pick a number 111 db SINAD and dynamic range (equal to a $9 Apple USB-C dongle). 32 bit fixed with some processing probably won't effect real world 24 bit results, but one might need to be careful with some of the DSP. Floating processing one has tons of range to work with.

Perhaps @Lonnie could explain further if I've overlooked why AVP's can't manage better performance as basic DACs.

 

[GrimSurfer]

As part of @Lonnie's response (and apologies if I missed it) could he tell us what department he works for at Emotiva? Engineering, Marketing or other?

This way we can frame our questions accordingly, so as not to put him in too difficult a (technical or policy) position.

 

[DonH56]

I think we are in agreement. The various multi-tones I used covered a range of 100 db. I dropped them down to EQ, and then sum before bringing them back up. I of course don't care about any signals 500 db apart in size. My point is that such processing done in 32 float won't prevent an AVP from achieving just to pick a number 111 db SINAD and dynamic range (equal to a $9 Apple USB-C dongle). 32 bit fixed with some processing probably won't effect real world 24 bit results, but one might need to be careful with some of the DSP. Floating processing one has tons of range to work with.

Perhaps @Lonnie could explain further if I've overlooked why AVP's can't manage better performance as basic DACs.

Yah, I think so, and think we always were, just tossing terms around. I do not see how any reasonable DSP can limit the performance to such an extent (comparing stand-alone DACs to AVRs/AVPs in general, not just Emotiva's products) and suspect noise coupling and other issues delineated earlier are the culprits. Whether the 32-bit math is integer or floating-point is in the mud unless something is really bonkers in the processing -- at least IME.

Note that whatever math is used, if you apply 20 or 30 dB attenuation to offset the summing of signals, for EQ, or whatever, the resultant signal applied to the DAC itself may effectively reduce SNR. To handle the peaks the average level in that situation is reduced by several bits. I suspect that is what Lonnie means.

 

[GrimSurfer]

Yah, I think so, and think we always were, just tossing terms around. I do not see how any reasonable DSP can limit the performance to such an extent (comparing stand-alone DACs to AVRs/AVPs in general, not just Emotiva's products) and suspect noise coupling and other issues delineated earlier are the culprits. Whether the 32-bit math is integer or floating-point is in the mud unless something is really bonkers in the processing -- at least IME.

I suspect noise coupling is the most common issue...