Living Room System

Wednesday, February 18, 2015

Shades of Tweakdom

There's a spectrum of potential value in audio improvements and "tweaks."

1.  At the most important end of the spectrum are things that are well understand and known to academic audio engineers, such as frequency response and audible reflections, which often have easily measured effects.  In this area, there are well known things that can be done to get better sound…better speakers, better designed room--including speaker and listening position placement, room mode and reflection absorbers, etc.  Many of these effects can be easily measured…but wrt absorbing room modes, it takes far more absorption (such as filling all the corners out 3 feet) than most people would imagine or tolerate.  Meanwhile, the effect of a tiny bit of absorption might be hard to measure simply because it doesn't have much effect.  Also, contrarians may dispute whether improvements are actually being made--in which case we have kind of a fashion competition, with subjective preferences ruling.  Do you like a more or less lively sound?  In any real room, you will be getting some reflection, and reflection isn't fidelity, but some reflectiveness may make for a more enjoyable experience anyway.  Most important is the quality of the reflected sound--how much does it change the frequency content of the original sound?  And there you have tradeoffs such as trading off the frequency balance of reflection for the total amount of reflection.

2.  Then there are things that have measurable differences that are nearly within the normal audible range, but are generally believed to be below the threshold of audibility for various reasons.  For example, distortion below 0.1% is plausibly inaudible, but many seek amplifiers with PPM or less distortion.  All other things being equal, lower distortion would be better.  All other things, unfortunately, are never equal, high order harmonics may be created by attempting to cancel low orders, etc.  Frequency response to 30 or 50kHz may be in this category as well--it is just a short distance, say a doubling or tripling of frequency, from where people have known hearing ability.

3.  Then there are things which might be measured, such gigahertz frequency response, but have very implausible relationship to human listening.  Many high end cables have alleged benefits in this kind of area.

4.  Then there are things that have a plausible explanation (still plausible to a very open minded scientist, for example) but can't easily be measured.  Here I could think of the benefits of PLL over pulse count detectors.  My information loss theory regarding DS decoders may be hard to measure, certainly it can't be measured with standard technical instruments.

5.  Same as 4., but theoretically impossible to measure the effect.

6.  Things that have an explanation, but not plausible to most serious audio scientists.  For example, quantum field generators.

7.  Things that have no explanation, and are only sold on the words "listen for yourself."  These things may be pure con.  Given the high variability of listening, the placebo effect, and more, many people may perceive a worthwhile change even when nothing useful is done.

"Information" is not just my imagination

When I say that PCM systems preserve information better than Delta Sigma systems, I mean it.  I'm not only talking about my subjective experience (and I don't have a strong belief in subjective experiences, I believe we're easily fooled and fooled all the time).  I'm talking about something which I strongly believe could be measured, in principle.

One type of measurement would be to create a modulation transmission system using a digital system.  Take two 56k modems, connect one to the input of the digital sampler, and the other to the delta sigma output.  See the maximum rate of information that can be transferred.

Now a 56k modem would not be a good test as it doesn't transmit as much information as a 20kHz bandwidth system could. It would be only testing the middle frequencies available, in which delta sigma systems might be superior to some PCM systems (say, DSD vs 44.1kHz PCM).  So imagine an even better modem, a 20-20kHz modem, that perhaps transmits data at 1Mbps, I don't actually know what it could potentially do.  But such a modulation transmission system would, I believe, show PCM systems to be superior.

Meanwhile, even a 56k modem could show that low bit rate MP3 loses lots of information.  Obviously a modem stream riding on MP3 can't transmit a higher bitrate than the MP3 itself.  We are audibly fooled because of limitations in our ability to hear, but the modem isn't fooled.

With an even higher bandwidth transmission test, say 20-50kHz, high resolution PCM would be far superior to 1x DSD.  While DSD nominally has frequency response to 50kHz, it's very noisy response.

In a way, you could see the issue simply in the rising noise above 10kHz.  But static noise does not perfectly correspond to information loss because there is also dynamic noise, modulation noise, and THAT is the big problem with delta sigma systems.

Just looking at the decoder, a delta sigma or sigma delta DAC seems to have excellent amplitude characteristics.  It can be made with higher apparent linearity than a PCM DAC.  But it achieves this by smearing the time response.  You have a feedback system that takes a variable amount of time to stabilize the amplitude level, or a feedforward system based on the addition of known quantities--either way information is smeared.  True PCM systems put out the best approximation to an analog voltage level at predetermined times, times that don't vary in a way correlated with the amplitude level.  The latter preserves actual information better because any kind of correlation smears information.

I'm pretty much flying by my intuition in describing the whole problem, and I could be wrong here, but it makes sense to me.  Meanwhile, there's also a problem in that it's not clear humans can perceive the higher rate of information.  My argument remains that it may not make a difference in a single listening session, but over the course of a lifetime of listening there will be more different listening experiences because of greater information available.  Such differences would be nearly impossible, effectively impossible to produce DBT proof for.

High Resolution is not Fake Snake Oil

High resolution PCM recording is measurably better.  There is not a myth, you can look at the actual measurements of real equipment using the best test equipment in many Stereophile reviews by John Atkinson, who has published detailed measurements of more equipment than anyone, over three decades of reviewing and editing the most popular high end audio magazine.

The question is, can you hear the difference.  Some say yes, including many industry professionals who are engineering graduates, or the top recording and mastering engineers who have won grammy awards, many reviewers, many high end audio equipment buyers.

But none of those people have the certified published and replicated DBT results that show that there is an audible difference.  At least this is the belief of mainstream academic audio engineering, also the likes of Hydrogen Audio and the Boston Audio Society, and it is verified in a critical scan of published articles in the Journal of the Audio Engineering Society.

Now, in many cases the Actual Mastering done on a high resolution recording is different, including the possible use of less limiting and compression, lower average level, and so on.  It would seem to me it would be worth the bother of high resolution PCM to get that.  Maybe others think it is a scam which they'll boycott.

But anyway, this is not tiny rocks, or crystalline wire compounds, or cryogenically treated AC outlets--such things as have no measurement even remotely relevant to audio measurements.

The same would largely be true of special cables, at least the ones that don't actually distort the sound in some way.  Measurements are essentially invisible to the kind of standard tests John Atkinson performs.   Sometimes cable differences are there if you perform the right measurements, say at gigahertz frequencies and so on that wouldn't be measured by John Atkinson.  That would be true of the principled cable solutions from Cardas…based on litz wire same as used in very high frequency instrumentation.  There ARE measurable differences, but not on standard tests, on frequencies way outside normal ranges.  So cable differences of THIS type are in sort of a middle ground.  Most cable differences are pure hype based on construction or imagination rather than measurable differences.*  Dialectric absorption of different dielectrics can be measured, but then with a low frequency exception to their being measured in conventional tests.  But most cable differences are hype, or at least people could get nearly all the benefit from a professional grade cable with polyethylene rather than vinyl dielectric.

(*Same is true of many other audio tweaks, the crystals, quantum field generators, cryogenic treatment, and so on.  Though separated wires is possibly measurably better if not actually audibly better.)

So Just Thought I'd Point This Out, to the serious people decrying high resolution recorded audio.  It is measurably better.  All evidence suggests that if it is ever proven to be audibly different, it would be audibly different and better.  At the same time, it has no support form serious evidence (at least that hasn't subsequently proven unreplicable) that it is audibly better.

That's now and in general.  If suddenly my hearing gets much better, I could hear it easily.

But most important may be the argument I've made elsewhere, which is that even if differences can't be reliably discerned, there may be a wider space of possible experience which would become evident over a large number of listenings.  The wider space of possible experiences is made possible by there being more information.  Supposedly inaudible frequencies nevertheless apply variance in this way to the experience space.

So it's not in the same category as many audiophool things.  It's provably better, just hasn't been proven for being audibly better.  (Many people insist it is.)

I think it's a good idea at 24/96.  But I might draw the line at some point.  Maybe.

Saturday, February 14, 2015

Acoustat vs Sub Impulse Response
Above is the graph made by Tact when I was time aligning an Acoustat panel with an SVS sub.

Notably they both initially seem to respond out-of-polarity initially, and the sub for a longer period.  But taken as a whole, the impulse has correct polarity in both cases.  I have in other ways verified that the Acoustat polarity is correct, such as with an old "SoftPolarityTest" signal, which I can view on Android oscilloscope.  The sum total of an audio signal is correct, but a tiny leading transient (exaggerated in the Tact HF measurement shown above) is out-of-polarity.

Today I added one more verification, using the Android PolarityChecker app.  I copied the mp3 files from the phone storage (where the App puts them) onto my Mac, and then updated my Sonos libraries. Then I was playing the polarity test in living room and bedroom.  Both systems show correct polarity, and the Acoustats individually.  This is clearest full range, where I get the green signal more than 4 times out of 5.  With the 4kHz test, about 1/3 of the time it shows incorrect polarity.  I believe this is because, as the plot above suggests, there is significant phase shift, sufficient to cause an effective out-of-polarity initial transient at the higher frequencies, or at least look that way to a test comparator a significant fraction of the time.  The acoustic transient response shown above suggests serious phase shift above 10kHz, perhaps as low as 3kHz.  But this is also where audibility of phase shift is lowest.

Likewise the sub has phase shift in it's initial response due to roll off above 300 Hz in the speaker itself, and 80 Hz in the very steep crossover.  That could explain it's long initial out-of-polarity component.

My guess for now is that the Tact image is roughly correct.  But it has also occurred to me that the picture shown might be of the Tact transient itself, which might have a significant leading out-of-polarity component.  That was why I wanted to resurrect my old LAUD 3 measurement tool, which computes a very clean transient response (from a MLS signal).  But so far that effort has been hampered by the old computer with ISA bus that it requires for the vintage DSP card from Turtle Beach.  Some time ago the Fiji card was removed, perhaps it had issues.  I tried installing a spare Pinnacle card, but the floppy drive wasn't reading the drivers.  I later found I had disabled the floppy drive because it was stuck and preventing the ATX power supply from ever turning off.  So it needs a new floppy, or I could burn the drivers to a CDROM, which I may do soon.

Sunday, January 11, 2015

Current Crossover Settings

As of January 10, 2015

Behringer DEQ (used for Acoustat high pass and EQ)
  Parametric EQ's
    80.5 Hz, LC, -15.0dB; 1/2 of LR24
    80.5 Hz, LC, -15.0dB; 1/2 of LR24
    227 Hz, 1Octave,+3dB; sweetener Eq added 1/4/2015
  Gain Offset 0dB
  Delay 4.14 msec Left and Right

Behringer DCX (Used for subs and super tweeters)
  SW Left
    A -14.1dB
    LR24, 80 Hz
      1. 20 Hz,  +1.0dB, Q3.5; room curve boost
      2. 45 Hz,  -7.0dB, Q4.0; #1 room mode
      3. 39 Hz,  -3.6dB, Q5.6; #2 room mode
      4. 32 Hz, +4.0dB, Q4.0; room curve boost
      5. 68 Hz,  -3.0dB, Q3.2; #3 room mode
      6. 20 Hz,  -3.0dB, Q1.0; Tilt down to cancel excess up tilt
    Delay 0.1msec

  SW Right (Gain pot straight up)
    B -14.1dB
    LR24, 80 Hz
      1. 27 Hz,     0dB, Q1.6; old boost discontinued
      2. 45 Hz, -5.0dB, Q5.0; #1 room mode
      3. 66 Hz, -3.0dB, Q3.2; #3 room mode
      4. 32 Hz, +3.0dB,Q2.0; room curve boost
      5. 25 Hz,      0dB,Q1.0; old boost discontinued
    Delay 1.15msec+0.29msec

  ST Left
     A -2.0dB
     Delay 2.18ms+1.31ms

  ST Right
    B -2.0dB
    Delay 2.56ms+1.02ms

All Polarities are NORMAL.

Both Subs fed through XLR 10dB attenuator.

N.B.  Cutting Subwoofer level is difficult; SVS's are high sensitivity compared with Behringer DCX which has high output, even with 10dB attenuator.  "gain pot straight up" is at least -3dB, which may be what is dialed in other side but not sure.  Experimental cut to SW Left will require adjusting SVS input level, fortunately it's a digital level control.  With level adjust, EQ 6 may go away.

Currently listening to my altMusic folder random shuffle through Sonos, taking a break from Pandora and William Orbit.  Beatles playing right now.

Bedroom System

    -13.7dB Sum
    80 Hz LR24 LP
    Delay 0
    EQ 50 Hz -4.0dB Q4.0; Room Mode
    -1.9dB A or B
    80 Hz LR24 HP
    Delay 3.49 ms
    EQ 107 Hz +4 Q3.5; sweetener added 1/4/2015 re floor/ceiling cancellation

Friday, January 9, 2015

Music is experienced constructively

One of the things I was taught in Cognitive Psychology is that all perception is constructed.  We directly sense various nuances…and construct those into complete experiences.

This is true for visual images…and it is even more true for Music, which doesn't happen at once but over time, requiring memory.  And the unique feelings that Music evokes…are of course constructed from the harmonies, rhythms, and more complex relationships.  Most of the ultimate experience comes from the mind-as-it-exists-at-the-time, which itself changes from one listening to a piece more than once.  Your memory has already grown because of the first listening by the time you start the second.  But still, most comes from many previous experiences.

So a little bit of information, taken from a very information rich source such as 16 bit PCM, is sampled, and that little bit of information is turned into far more virtual information in the complete experience…that would swamp the largest computer clusters to simulate.

The variability in experience from the same CD comes from many levels…how we choose to sample the original, and how we choose to expand on it to be come an experience.

We sample far less information than is on a CD (making sampling differences possible) but that doesn't necessarily mean that a higher resolution source, having more information still, wouldn't sample sufficiently differently to be detectable as such.

In every set of samples there are common elements.  The larger the ultimate information space, the more diffuse those common elements, and therefore the richer the possible experiences.

If information is perceivable, more information ultimately makes for noticeably richer experience the second and all subsequent times.  (This might not argue for higher sampling rates…but it would argue against all lossy compression systems.)

At the same time, much of the experience comes from state of mind, attention, need.  This is more stuff that is not really part of the recorded information.

State of mind is highly affected by belief, such as belief in strategically placed crystal resonators.  If you have the belief, you will of course experience that you need those crystals, and be sure that you need them when you don't.  And you wouldn't be lying if you said your experience was affected by them, even if they had negligible audio effect, if you had another way of knowing if they were in use.

Belief may also cause you to seek out confirmatory examples, and find alternative explanations for examples that don't confirm.

Tuesday, January 6, 2015

Figuring out the Notch

As I posted a few days ago, after the most recent adjustments, my left channel response (uncalibrated in this graph) looks like this:

Generally it follows a nice room curve, showing about 12dB rise at 20 Hz, which is the peak frequency, with useable response down to 10 Hz.  The high frequency shown is 250 Hz, which is in a slight depression about 4dB below the true baseline at mid and high frequencies.  Generally it's relatively smooth compared with the raw response (full of severe room modes, btw).  Right now what bothers me most is the notch at 80 Hz, which got slightly worse as I tilted the response down 4dB at 20Hz (the 20Hz peak was 4dB higher before the adjustment).

The notch looks as though the high pass and low pass got separated somehow.  But they are both set to exactly 80 Hz, and the time delay between subs and panels has been corrected also.  Although, I wondered about the cumulative effect of other time constants in the system response, and so I tried changing the crossover setting on the high side only from 80Hz down to 70Hz, and then down to 50Hz.  At 70 Hz the response was only slightly different.  At 50 Hz, the notch got worse and seemed to  create a new hill around itself (although, even at 80 Hz, you can see a bit of a hill around the 80 Hz notch).
Judging how that went, I began to seriously wonder if separating the crossover frequencies would work better--the opposite of my original visual intuition.  But what I measured instead was the response of the Acoustat left speaker by itself.  That looked like this:

The Acoustat by itself (in current location in my room…which is probably the causative factor) has a series of saw like patterns in the penultimate bottom end.  These may be reflective cancellations, but the back wave cancelation would be the small peak at about 140 Hz…which is strangely a peak instead of a cancellation precisely because the Acoustat is bipolar…the back wave is already 180 degrees out of phase with the front.  But then why are the peaks lower than 140 Hz in increasing size, until we get to just above 70 Hz?  I can't explain the peaks below 140 except to suggest they are also reflections, possibly involving the entire room or the hallway.

The 70 Hz peak in the Acoustat response could be precisely what appears in the combined response (top) which also has a peak at 70 Hz, the left side of the notch.  The notch cut at 80 Hz in the combined response near the notch cut in the Acoust response just below 90 Hz.  The Acoustat response rises oboe that, possibly contributing to the mountain in the combined response.

The notchy Acoustat response doesn't exactly explain the singular pronounced notch in the combined response, but it could be a big contributing factor.  It might be worthwhile to cross the Acoustat over higher with, perhaps 100 Hz (at the peak in Acoustat response) rather than 80 (deep into the notch valley).  From that vantage, the crossover is working with the response error to cut response below 100 Hz, and the falloff below 100 Hz in the Acoustat response will be slightly filled in, but beyond that slightly removed, leading to a smoother overall curve.

It could be I also need to reduce the bass level rather than apply an overall tilt in opposite direction to room curve (reducing room curve boost by 20% or so across the board) as I am doing since Saturday.  Reducing the level would lead to less undesirable reinforcement (the unwanted bloom) in the 70-150 Hz area.  Reducing the level 4dB would have the same effect as the tilt on 20 Hz, but additional reduction for higher frequencies up to the crossover point.  (It's hard not to see this suggested level change as "less bass" even compared with the tilt.)

Additional note: the deepest notch occurs at 280 Hz.  That has a wavelength of about 5 feet.  At the wall reflection polarity is retained.  The back wave starts out-of-polarity, so as it goes through a full wavelength in reflection, it cancels maximally.  That suggests a wall reflection is occurring at 2.5 feet, and indeed that's about how far my Acoustats measure from the wall perpendicularly.  On the listening axis, it's more than that, and that could explain the dip at 180 Hz.

Thanks to angling and other factors, these reflection notches are not as bad as they could be.  I think at minimum about 3db Eq with bandwidth of about 1-1.5 octaves could be used here, with center frequency of 230 Hz.   Reflection and/or aborption treatments?  I don't know, maybe they could help…but I'd be afraid of them hurting.  Moving the speaker much is basically impossible.  I chose this 2.5 feet from wall (it seems more like 3 feet) as the best possible in a multipurpose room.   As it is, the room is already squarely in man cave territory.  I have friends and parties, but I live by myself now, and my friend has accepted the speaker portion of the living room (but wants changes elsewhere).  Speakers 4 feet from the wall would not work with my schedule of monthly parties.  All that speaker moving…which needs to be done to the nearest 0.1 inch or better, I'd go nuts.