Saturday, October 27, 2018

Emotiva headroom and distortion measurements (new unit)

Emotiva lists 1.0V as the standard output for the XSP-1, however, the distortion is actually spec'd at 2.0V.  No other voltages are listed.  I was reassured by Emotiva that "6 Volts" is the maximum, but I wasn't entirely clear on whether this was input or output, and I'm concerned about the 3.8V or so from the singled ended output of my DVD-9000 playing HDCD's.

I've measured the maximum voltages now, somewhat approximately because I'm only using "instrumentation loads" like my 10M DVM and 1M oscilloscope.  This is a factor that 10 years ago and before I simply ignored.

Anyway, for single ended I/O the maximum input is very impressive at 9.6V RMS (measured with the gain at -20dB).  Maximum SE output is around 6.8V (gain at +12.0) at onset of obvious clipping, a tad bit of obtuse rounding can be seen above 5.5V (looking on my newest digital oscilloscope).

Driving the output into very hard clipping close to 10V can cause square wave doubling.  This is an undesireable per se, but in most circumstances your home amplifier will already have exploded at 10V sine wave input.  This is not an instability, what happens is that after the top of the sine wave flattens, the center begins to grow negatively proportional to further drive, until this center downward spike reaches the bottom, making the clipping wave, now pretty much square, into a doubled square wave at twice the original frequency.  It shows essentially that multiple internal elements are clipping at not exactly the same time...therefore something like a push pull output.

True pro equipment can handle 10V as a standard level.  That's probably why Emotive lists the standard level as 1V, to make it clear it's not 10V pro gear--which is compromised in S/N at lower consumer levels.

These I will not be pushing the Emotiva limits in any way, though a recent adjustment to the Lavry was to set max level (-8dB) to -17dBU which is 5.4V, which would be a touch too high for single ended output on the Emotiva, but probably gives around 6dB headroom for balanced output (not measured yet, that presents some challenges).

Certainly the input will handle anything I can throw at it input wise.

On my Juli@ card, using RMAA, with balanced I/O last weekend I measured about 0.0005% or 0.0006%, with my residual at 0.0003%, meaning the actual distortion is around 0.0003%, as shown in website graphics (showing better than the claimed spec of 0.0005).  These are measured at the 4V input level of the card, minus about 0.9dB.  I measure this either to 96kHz sampling or 44.1kHz sampling with the same result, showing no super high harmonics, very clean top end.  Basically my residual and the Emotive line up perfectly, save a little more 3rd harmonic (the push-pull output cancelling any 2nd order distortion) and down below -120dB a tiny bit of 7th (generally, stuff below -110dB can be ignored, the Denon DVD-9000 is full of high frequency hash at -115dB but is one of the best sounding CD players I've ever heard).  It was this cleanness that sold me on the Emotiva, among other things, and having the best phono circuit in my collection clinched it as my living room analog preamp, requiring me to get a second for the bedroom I had originally purchased my first unit for several years ago.  The THD levels compare favorably with the BEST current Mark Levinson at $40k, and the Emotiva has the essential-for-me digital volume control if not as nicely implemented as the Levinson.

On my Sound Technology, I get measurements showing that single ended output should be kept below 4V.  (I actually don't use the single ended output for anything except driving the Sonos box at 2V in the bedroom.  The balanced outputs go to the Masterlink in the bedroom, and the Lavry in the Living Room, my two most critical things.)

6V output...4% THD
5V output...0.38% THD
4V output...0.0063% THD
3V output...0.0046% THD (this is about my residual on this analyzer when it's working great, once in a blue moon it can hit 0.0044%, but right after taking these measurements it decided to go beyond 3% resdual, a sign of internal overwarming or something, so I shut it down).

In balanced output, this would probably translate to a maximum low distortion output of 8V, well more than what I need for the Masterlink (4V) or the Lavry (5.4V).  Right now, with cranky Sound Technology analyzer, I can't do the measurement I had hoped to do this afternoon to confirm that.

Bless the Sound Technology, after 3 hours rest it came back on and I was able to record a better residual, which is I think not untypical actually (until after a few hours, it will go up, I adjusted my ST that way so I could do decent measurements without 3 hours warmup, I'm usually just doing a few measurements, though this way, in 3 hours you can't use it optimally).  That is:

Sound Technology Residual: 0.0033% THD

Keep those in mind for the above measurments, the 3V Emotiva output is not exactly the residual, it's 0.0013% higher, but that might depend on the number of minutes of warm up also not that much.  Actually I know from other measurements 2V is at 0.0003%, so the 0.0013% probably says more about the ST and it's state of warmup (remember, it failed just a few minutes later, and I had to let it cool down for 3 hours).

Anyway, with some problematic adapters they may have been contributing, I measured output distortion balanced, once again using the ST as primary source.

At 8V balanced, distortion was a hair above the residual at 0.0038%.

At 10V, distortion had risen to 0.03%.

At 9V, distorition was also around 0.03%

So my prediction about 8V being the highest low distortion output level seems correct, the peak optimal RMS levels are:

4V single ended
8V Balanced

At these output levels, THD has not increased since the lowest output levels, and may even have reached its lowest level.

It goes higher, but distortion starts rising, fwiw, still way below speakers and such a few dB higher.  I'd keep it below 6V single ended and 12V balanced in all cases with great prejudice, though it doesn't really break up till even higher.

I could even crank up the input level on the Lavry, now set to 5.4V, a few dB higher.









The Kitchen HT system sub now re-enabled for stereo

I mainly use my kitchen HT system to play stereo, but I have the two surrounds and sub to add in for actual 5 channel material.  I'm opposed to center channel speakers on various grounds, and I always use the appropriate setup which folds the center as a mono signal into the fromt left and right.

For stereo, I'd always used the "Direct Stereo" output of my Yahama HT-5790.  I measured this, and it clearly bypasses the digital conversions that the various HT modes require.  An alternate is "2 channel stereo" which does do digital processing for 2 channels.  The digital processing is clean except, of course, it introduces noticeable pre and post ringing, the kind that J Peter Moncrieff seems to be telling us is all fine and good.  (And all "objectivist" engineers also, I might add.)

So, using Direct Stereo, I have no "subwoofer" output, because the crossover is implemented in the DSP.

Recently I've been playing FM on the Kitchen Tuner because it attached to my roof level whip antenna which clearly surpasses all my other antennas (merely as a result of height, long I tested inside and it was worse than my indoor antennas).

After I moved on from the McIntosh MR78 with Modafferi Mod, which had a kind of steely, though clearly very high information, sound that was fascinating sometimes but not appealing and actually a bit tiring, I brought out the Sony 730ES (because it was most accessible in my conditioned storage building) and the sound is so much "nicer" somehow, combined with the much better antenna, it's somewhat intoxicating.

But what's always been noticeable was the kind of "miniaturization" done by the Kitchen setup.  This may be partly the almost nearfield location of the speakers, and similar factors.  The speakers themselves are the once Stereophile Class A Revel M20's, which go to 30 something Hz and cross over nicely at 60Hz, as I have always done in the bedroom.

I've always noticed the "heft" of the living room system playing FM, so preferentially even I play the Living Room stereo when I'm in the kitchen, even at the expense of an actual stereo image.  Appropriate somewhat to FM, it was the "concert is in the next room" experience, but like a real concert, instead of the table-top miniature in the kitchen.  Having the subwoofer for regular stereo, now adjusted pretty good, gives things the full size and heft, and even seems to be removing some of the steelyness from the tone--which might have been resulting from mismatch between extended high end and rolled off low end.  It may have been that lack of steelyness which I was confusing for the lack of definition in 2 channel (dsp) mode, much more than the digital resampling and DSP.

At various times I've played the subs, and though this speaker (an SVS 10" from the 2000's which was my very first "real" subwoofer and was a revolutionary upgrade to my bedroom system, replacing the McIntosh M22's I had been using as subs, but that was quickly upgraded to the SVS 39 inch 16 Hz tube). but in the kitchen it had sounded terrible, quickly invoking the "turn it off" reflex, so I guess long ago I had decided Direct Stereo was better and that was that.

But, I wondered, what if I just turned down the subs a little.  I had never really measured the system, or adjusted by measurement, or maybe I'd tried the auto adjust once and hated it, preferring my own manual "by ear" adjustments, notably using the built in level adjust test signals in the receiver, and balancing the different speakers and the subs by ear listening to the test signal.  This was set up before I had an always ready 1/6 octave RTA in my phone.

So I tried adjusting the sub level to get the flattest response using my standard noise source, Stereophile Test Disc 2.

And this required an astonishingly low setting for the subwoofer level.  After several days of readjustment and measurement and listening I've gotten to setting the subs at the low end of the second lowest mark on the level screen, with a 60 Hz crossover and "bass" set to "both".  As I was previously attempting to audibly match the level of the 1kHz centered pink noise signal with the signal in the bass on the Yamaha, I was setting to within the third highest mark.

If I set the "bass" to "swfr" it seems that the low bass disappears in "2 channel mode".  Im suspecting this is a bug in the Yamaha.  That's why I'm using "both," which I assumed would effectively use the bass as fill-in, and since I've set the back speakers to "small" they shouldn't get any bass from the "both".

What I really plan to do is use a behringer DSP to do the fill-in crossover and shaping for the bass, and I currently have a spare DEQ 2496 which I'm clearing the space for now.  Previously, way back in 2009 or so in fact, I bought an extra Behringer of the earlier vintage, DEQ 8020 or something, only 20 bit, and it was much more painful to use having few buttons and unintuitive to me displays, so the "equalizing the kitchen sub" project never got off the ground because I hated the equalizer.  Now I have a spare 2496 which I like (and will have to get another spare if I decide to continue this, or perhaps another miniDSP).

Now what I plan to do run the main speakers straight through, perhaps in Direct Stereo, and capture the extra LR line outputs in the back for the DEQ, which will then perform crossover, eq, level, and perhaps other functions.  However, I'll have to have a switched connection to the sub outs for multichannel inputs, which simply go straight through the receiver

Moncrieff and Linkwitz

J. Peter Moncrieff, editor of International Audio Review, had been quiet for a long time, his website having last been edited in 1999.  I tried to buy back issues in 2013 without success.

But now he's reappeared, with an incredibly long winded (as usual, and I haven't read it all) takedown of MQA, and not just MQA but similar supposedly transient improving techniques in digital audio (which he calls a "modern revisionist digital engineering movement worldwide").


A lot of what Moncrieff is saying is correct.  Sure, he's making mountains out of molehills, exaggerating to the max, and full of puffery and self-congratulation (which I find entertaining, actually, like listening to PT Barnum, but now if he could only be a little less repetitious it would be less tedious).  But, still, much of it is correct, and I believe MANY wise old fashioned engineers would agree with the thesis that modern short transient digital filters that seek to eliminate pre and post ringing, including MQA, are WRONG at least in principle, pretty much along the lines Moncrieff is saying (though I'm not sure all his thought experiments are precisely correct, they capture the jest in a very accessible way, and his gift at doing that makes me find it worthwhile to read).  The old fashioned digital filters take full advantage of the sampling theorem's promises to capture all the information in the bandwidth window, up to the limit of their technical features (such as the number of times of oversampling...which does benefit from greater than 8 times, just as Moncrieff says, though many old fashioned engineers would say 8 times is plenty good enough) and the new kind not so much--the new kind ARE information lossy, with MQA being the king of the hill in lossiness of anything that claims to be CD quality or better.  If these new digital filters have benefits at all, it's through the euphonic effects Moncrieff describes, not actual higher fidelity.  It's like NTSC color TV's from the 1960's, first they up the color temperature to 9000K to make it look brighter, then they add nonlinear red push to make the skin look natural regardless of that, then they peak the horizontal response to look like there is more sharpness than there actually is, never mind the false edges and other artifacts.  A properly designed TV of the same basic performance would look less bright and less sharp, but would be more accurate and contain more real unobscured information to the serious viewer.  A properly designed digital system may sound duller and less spacious than MQA, but it's true hifi and not euphonia.  Though I'm also of the opinion the difference between MQA and regular PCM would be very hard to hear above the level of chance.  Still, and then perhaps even moreso, why not have true high fidelity?  And especially if you have to pay more to a middleman to have the fake.  Perhaps MQA should be understood more as a watermarking system than a high resolution audio system.  And the same is true of SACD and DSD-64, not that many care anymore.  (I keep my vintage Sony 9000ES, with true 1-bit converter at 10x oversampling, on the grounds that to enjoy the sound the producer intended, you need to use the matching decoder, and many SACD's do indeed sound pretty good for good production reasons and despite the inherent lossiness of the system.  The same is true of HDCD, which I think better than SACD, and has zero information lossiness whether decoded or not, but is dynamically lossy if not decoded.)




On another topic, one great audio engineer passed away recently was Siegfried Linkwitz, the primary inventor and promoter of the Linkwitz Riley crossover now used by many manufacturers and builders (including me, since about 1983),  the designer of many great DIY designs, and creator of a website with a vast amount of incredibly detailed audio  information and analysis. 



This was strangely ironic for me, as about the same time as Linkwitz passing I was discussing crossovers with a friend who dismissed Linkwitz's claim that the group delay (phase shift) caused by properly implemented LR crossovers is not audible, or is at least not audible to him.  My friend described Linkwitz very negatively regarding this claim.  I believe Linkwitz was honest and a careful listener and these differences ARE hard to hear.  Here is Linkwitz' page on the subject:


But I've long considered the idea of transient perfect crossovers to be appealing, so I've launched a new set of projects to try them, using miniDSP processors and FIR digital filters.  There will be a steep learning curve in this.  (Linkwitz has other pages where he details the problems in "phase perfect" approaches.)  Ask me in about a year if I got anything working.

Friday, October 19, 2018

MiniDSP opener

All these distractions from my #1 audio concept: linear phase DSP.

Wow, I can actually run rePhase 1.3.0 on my Mac using the latest Wine from Wine HQ!

Loading in the number of taps, I was reminded of the 48kHz sampling rate limitation of the OpenDRC 2x2 plug in for my OpenDRC-DI.  With my 40kHz super tweeters and all, I wanted to stick with 96kHz, which is what I resample analog sources to.

Well apparently the sampling rate can be boosted to 96kHz using a different "plug in."  I'm beginning to see how this works.  The OpenDRC-DI can accept digital inputs up to 216kHz, but the get asynchronously resampled to whatever the plug-in runs at.  And if the plug in runs at a higher sampling rate, then there are fewer taps.  This IS an issue for the ultimate DSP control, especially at low frequencies where you want a lot of taps to deal with complicated room issues.

This thread discusses the possibility of running the miniSharc 4x8 plug in on the OpenDRC platform to get 96 kHz capability.  The downside is that the taps are reduced to 2048.  Using the standard OpenDRC 2x2 plug in, 6144 taps are available.

It seems to me I could use the 4x8 plug-in on my supertweeters where the 96kHz rate is useful, and then run the 2x2 plug in for the panels and subs.

One nice feature in this regards is that the miniDSP's are going to be putting out the designated sampling rate no matter what the input source.  So, I will only have to do the time alignment between DACs for one sampling rate, instead of separately for each sampling rate.  In other words, I just do the time alignment once, and it's set, regardless of sampling rates and the fact that I'm using a different sampling rate for the super tweeters.



That Finnicky Linn Sondek

I've heard the advice: you are supposed to put your Linn Sondek on a small but rigid and light table...

Well, I'm sorry, my life is not like that, no extra floor space available for dedication to the turntable.

But I thought, I might get away with balancing the Linn by putting some fairly stiff small pieces of paper under the back feet.  I did that a couple weeks ago to a great feeling of satisfaction, seeing as how I'd made the turntable almost perfectly level, instead of the previous off by about 1 degree (or whatever 1 big mark means on my large circular level from KAB), tilted toward the back.

However, in infrequently playing discs since, I'd been noticing the Linn "isolation" was not what it was cracked up to be.

Finally, I put two and two together, and realized that my leveling mod may have increased instability. So I removed those little piles of paper behind the back feet, and presto, the fairly decent Linn mid-bass and up isolation has returned.  I can tap on the underneath the shelf pretty well without visibly upsetting the lp surface.  Previously, anything more that the very smallest touch would set it going.

The Linn guru Mark warned me that to level the table I should "fix" the stand.  He was right about that.  But it's not going to be easy to "fix the stand."  So it can wait until the stand gets moved back after the next remodel (replacing the window behind it has to get done before too long, and the floor underneath it).

Sadly the fixed Linn Sondek feet appear to be an essential part of the whole rig, and they must sit down atop a rigid surface, just as the Linnie advice suggests.  As far as the table being light, that another thing I don't believe and can't imagine accomplishing.

Also I don't believe in the superiority of not clamping, but for casual background listening, unclamped playing is far easier (one must remember to remove the felt washer!!!  a mistake I've made a few times) and works better on Linn than most tables.  It does increase the rhythm and tunefulness (Linn emphasizes the tunefulness, not the rhythm) however it blurs the detail--which is only really important for serious listening.  I find, however, that unclamped playing is slightly harmful for the tonality--which becomes a bit hashier even as it becomes more "tuneful."

The soft springs of the Linn actually protect the bearing when you are applying the clamp.  Only suspended tables do that, usually you are crushing down on the bearing when you apply the clamp, and that could be harmful to many turntable bearings.  On the Linn, the table only goes down a few mm before the platter hits the top of the plinth, which supports it nicely when you are clamping, with only the light springs loading the actual bearing.  I used to think it abominable to press the turntable down to the nice plinth top, but now it seems like it was made just for that.

The best way to clamp is kneeling down so the vinyl is at eye level.  Then, wriggle clamp down slightly for maximum vinyl flatness.  Some, but not excessive, pressure is required, mainly just enough to keep the clamp from slipping off the short Linn spindle.  Too much pressure may cause additional bowing with the edge rising.




Sunday, October 7, 2018

Hafler 9300

Now that the Krell FPB 300 shuts down in 10-90 minutes of operation or even idle with shorted xlr's and open outputs, I'm using the Hafler 9300 almost continuously.  The Hafler  a great amplifier, I believe, built with one of the world's greatest designs: the trans-nova.  I'm also lucky to have a minty and perfectly operating copy. I think the Krell amplifier might be a tad cleaner but after going back and rechecking, I've never been sure I could hear a difference, and many times I've mistaken one for the other, using thinking it's the Krell when it's really the Hafler.  This pertains to power as well, I've never had a case where conclusively the Hafler sounded less powerful.  Now I find that recent experiments were marred by volume adjustment creep on the DACs.  The correct adjustment is -7dB for the Krell and -4dB for the Hafler, a -3dB level difference.  I'm reminded of a friend who didn't believe levels could or should be matched.  My matching is within about 0.13dB, the best I can do and lucky it's that good, with the 0.25dB settings on my oldest Stealth DC.

Looking at the 9300 schematic, this is what stands out.

1) It's very simple for a reasonable power amplifier, though perhaps not as simple as I had been thinking, not simplicity uber alles.
2) It'd direct coupled with DC feedback and no servo.  The best when you can do it, and simple.  NO capacitors in the signal path to worry about.  No inductors neither, and no inductors or anything at the output.
3) The schematic is drawn wierdly around the supply capacitors.  There is nothing weird about the supply caps, in fact they are nicely bypassed with 4.75uF caps (probably film).  I had been thinking they formed a surrogate cap coupling, but not in this circuit.
4) What is a bit weird is that the - terminal is driven, and the + terminal is really the signal ground (explaining why the capacitors are drawn as they are...to show the ultimate signal ground).  The amplifier is internally inverting, but this is corrected simply by labeling the terminals in the correct polarity, therefore it appears to the user as a correctly polarized amplifier (just don't connect the grounds...which many amps warn you not to do when it wouldn't have mattered).
5) Essentially standard "the best" layout: input dual jfet diff amps with bipolar current mirror feeding bipolar drivers driving MOSFET output banks.
6) MOSFETs don't need no stinking current limiting, no stinking shutdown, no stinking anything except there are rail fuses, which have never blown.
7)The novel feature is the MOSFETs are in gain mode, making the amplifier inverting, so feedback is taken from their input.  I'm not sure why it is done this way, but I suspect ultimate "peak" power, which is actually the way it seemed on the bench.  Distortion began slowly rising above the rated level (150W at 0.003% distortion) however it seemed it might put out as much as 500 watts peak.  This power availability also translates to more than negligible output at 2 ohms or less.  Basically, there is minimal resistance (I don't see any!) in the output circuit, so it's wide open.
8) Feedback is therefore isolated from swings in power supply, and well as ultimate device linearity, meaning the amplifier doesn't "double down" when the stored power is running down.  That probably a GOOD thing.  It gives the best effort and moves on, rather than getting stuck in the mud.  Power supply isolation comes from the MOSFETs themselves, which are excellently so isolated.
9) Despite that lack of feedback around the output devices themselves, the distortion is incredibly low for a power amplifier.

I only think it would be slightly better with what the Krell has: a regulated power supply.  That where I imagine the differences I probably imagine the amplifier sounds as having comes from.

AND, looking at the 9303/9505 schematic, it's almost entirely different, far more complicated with many more bipolars in the front end and the jfet diff amp buried within all these bipolars.  Also there's a servo which the 9300/9500 don't need, and even...gasp...what looks like a protection cutoff circuit.  Not a jewel like the 9300 IMO--which doesn't need servo or protection circuit.  The Nelson Pass designed Adcom 5500 is somewhere in between, more complicated than the Hafler 9300 but less complicated than the 9303/9505, but, characteristically, uses almost entirely jfets and mosfets with just a bipolar driver stage (like the 9300).  While the Adcom looks like a fine amplifier, the Hafler 9300 is just the cat's meow of simple yet effective designs, I've never seen one done more nicely save the First Watt F5, which is power limited by comparison because there are just limits as to what's possible when leaving out the bipolar drivers.