Thoughts on bandwidth, sample rates and other miscellaneous spewing...

  • Note: To skip the miscellaneous spewing just jump to the last paragraph.

    In the pre-digital days of analog I embarked (with my colleague) on a challenge of attempting to capture the sound field of acoustic instruments or group of instruments (piano, harp, violin, viola, cello, double bass etc) in an acoustic space (room, hall, church etc) and reproducing that performance later.

    Studying the research of Alan Blumlein and others in the field of psychoacoustic and the reproduction of recorded sound using two loudspeakers (stereo) we experimented with various microphone techniques, equipment required to capture the sound, and then reproducing the performance in its acoustical environment . Critical listening of the transducers, electronics and microphone techniques led us to the conclusion that within the limitations of “stereo”, the Blumlein (X-Y) and/or Mid-Side (M-S) were the best compromises. In other words no mono multi-micing and pan pots.

    Determining the least compromised recording chain (microphone, mic preamp, storage medium etc) required extensive listening tests.
    Through a audio equipment review publication I had access to a vast number (fifty or more) of preamplifiers and power amplifiers and in a controlled listening environment these circuit designs were critically evaluated through extensive listening and measuring tests.

    The conclusion was that a properly designed circuit (preamplifier and amplifier) with “wide bandwidth” offered the most “accurate” ability to pass through and amplify a signal and convincingly reproduce the original performance in its acoustic space. Textures, clarity, attacks, overtones, imaging left to right, perceived depth and room acoustics were more convincing or “accurate” with wide bandwidth designed electronics.

    There are obviously a multitude of other factors such as negative feedback, transformers etc. that I won’t go into here. By the way, the recording chain that I settled on was completely tranformerless from microphone to storage medium. The end user/listener of the music of course used a 33 1/3 rpm phonograph disc so it was necessary to test that part of the chain via listening to the mastering of the disc process (half-speed mastering etc) and the vinyl itself (TELDEC ie Telefunken Decca virgin “pure”vinyl was vastly superior to anything. Hold it up to a light bulb and you can see light through it).

    A very talented electronics designer once told me that in order to accurately reproduce audio within the human hearing range requires a circuit bandwidth of ten times the highest reproduced frequency (20kHz x 10 = 200kHz). The same applies going below 20Hz. Although we can hear up to 20kHz why go beyond? Because in order to accurately reproduce up in that range the circuit bandwidth must exceed 20kHZ significantly. The engineer/designer words, not mine.

    Pulse tests, square wave test etc bear out his theory. Feed something in and watch if the circuit can accurately reproduce the impulse without overshoot and settle down again requires bandwidth. What does a 10kHZ square wave look like on the leading edge? Did not matter what the circuit topology was/is ie. transistor, IC, tubes, FET etc.

    Moving forward in time to digital capture and reproduce and try this simple experiment:

    Take a quality acoustic guitar (preferably a Martin made in the mid 1930s through 1944) that has responsive attack (impulse) and rich overtones and put it and the player in a nice sounding room.

    Set up a high quality (Schoeps etc) condenser pair of microphones in a stereo configuration such as “crossed figure eights at 90 degrees to one another”; cardioid or super cardioid at 120 degrees; figure eight and cardioid in the MS configuration.

    Record the performance at 44.1k 24 bit and at 48k 24 bit. Listen to the playback of each sample rate version over properly setup quality speakers or even quality headphones.

    What does this have to do with an electric guitar plugged into a limited bandwidth tube amplifier fed into a transformer and then into an even more limited bandwidth speaker (or that same signal chain only substitute the KPA after the electric guitar)?

    Accuracy is not a goal with most sound recordings. The Neumann U-47 or M-49 are considered by some to be the “best” vocal mics ever made and they were made sixty to seventy years ago. They really are not from a technical specification accurate but their euphonic colorations offer a desirable “quality” that many preferred both back in the days of Frank Sinatra and the Beatles through today even when thrown into the final end user’s listening experience ie. 128k (or less) lossy Mp3 and lastly the “in ear” “headphones”.

    Long way of saying:
    The KPA now allowing 48k (or higher) as the external master clock is a benefit to me because other acoustic instruments and vocals in the 48k 24 bit DAW project can/will sound better at the higher rate and the SPDIF out sounds better to my ears.

    Will

    Edited 2 times, last by WillB (April 4, 2018 at 12:34 AM).

  • Thanks for taking the time to relate your experience.

    I too 'feel' like the higher sample rates sound better, or feel better when I am trying to play and listen simultaneously.

    Kemper Powerhead w/remote & Kabinet
    Focusrite 18i8 (2nd Gen) - Windows 10 - Ableton Live - Yamaha HS-8's - DT770 80 ohms

  • To put it another way, FWIW

    Most of the records I make are recorded at 96k.
    Because that’s what sounds best to me.

    I’m not going to compromise that in order to be able to use the Kemper digitally.
    I’d rather the extra conversions.

  • There has been many proofs in the last decades that 48 and 96 kHz does not provide any benefit to the audio quality, compared to 44.1 kHz

    If the opposite was true, there would be scientific A/B audio tests available where the differences are easily demonstrated.
    Additionally, makers of reknowned audio equipment and software (that‘s where audio scientists go to work) that provides all sample rates, would recomment the use of higher sample rates for the benefit of their customers.

    I cannot find such A/B tests or recommendations.

  • That’s, with respect, nonsense.

    There haven’t been any “proofs” of any such thing.
    Only opinion.

    A B testing is only useful with people TRAINED as to what the artifacts are and how to listen for them.
    And also in truth it’s music we are talking about, not data.
    It doesn’t matter if an average person cannot identify a difference in a piece of music he doesn’t particularly know.
    But show a trained recording engineer or producer a comparison using a track he is intimately familiar with, and emotionally INVESTED in, and then he can spot the difference every time.

    I’ll stake my own reputation to no small degree... but when people as informed AND talented as George Massemburg talk about the value of recording at 192k it’s foolishness if you choose to ignore it because some armchair writers on the internets claim to “debunk” it.
    When I’ve heard their hit records I’ll take them as seriously.

    To your other comment:
    The makers of digital audio hardware and software DO make high sample rates available.
    One could as easily argue that if they were “unnecessary” then they’re wasting resources by offering it.

    People love to quote Lavry for his conclusion that sample rates above 60k might be unnecessary, but those same people almost never quote the other part of his conclusIon which was that lacking 60k conveyors, and given the current availability, he’d recommend 96k.

  • Take a quality acoustic guitar (preferably a Martin made in the mid 1930s through 1944) that has responsive attack (impulse) and rich overtones and put it and the player in a nice sounding room.

    Set up a high quality (Schoeps etc) condenser pair of microphones in a stereo configuration such as “crossed figure eights at 90 degrees to one another”; cardioid or super cardioid at 120 degrees; figure eight and cardioid in the MS configuration.

    Record the performance at 44.1k 24 bit and at 48k 24 bit. Listen to the playback of each sample rate version over properly setup quality speakers or even quality headphones


    Now here is the problem I have with opinions like these. Take an OLD name instrument (it’s gotta be old ‘cause, you know, old is better), and go on with a perfected list set up like a recipe and and THEN you have proof that it matters.

    No one ever does, and no one ever will so the «proof» is all anecdotal. And because of the list and requirements, if someone ever does it’s easily shot down: «oh, you used those Schoeps. THOSE aren’t the best ones for a 1930’s acoustic...». And besides, I wonder if real life is less than idealized (Ibanez acoustic from 2005 and Røde mics), then what? I mean... come on! :P

    I suggest an easier test for 44.1 vs 48/96/196/1000. It can be applied to everything and anything. Ready?

    «If it sounds good, it is good.»

    ;)

  • Dear Dahla,

    I mentioned those specific instruments ("Golden age" and "War era" Martins) not because they are old. I specifically mentioned that era and manufacturer, somewhat tongue in cheek, because the general consensus among musicians and luthiers is that those instruments represent some of the finest instruments ever made due to their materials, design, and the craftsman/artists who created them. I thought about using the example "Stradivarius violin" in my somewhat sarcastic example just so people with the "old" hang-up might give it a pass. I didn't. You didn't.

    For your benefit I've attempted to conform. How's this?

    ""Take a quality acoustic guitar (preferably it was made very recently with eco-friendly sustainable materials by a fully automated computer operated machine) that has responsive attack (impulse) and rich overtones and put it and the player in a nice sounding room.

    Set up a high quality (Shure SM-57 etc) pair of microphones in a stereo configuration such as “crossed figure eights at 90 degrees to one another”; cardioid or super cardioid at 120 degrees; figure eight and cardioid in the MS configuration.

    Record the performance at 44.1k 24 bit and at 48k 24 bit. Listen to the playback of each sample rate version over properly setup quality speakers or even quality headphones.""


    You said, "with a perfected list set up like a recipe..."

    If you read and comprehend what I actually said you will see I used the word "compromise" twice.
    Everything is a compromise...in case you haven't noticed yet.

    Finally:
    Did I miss something?

    You said, "I suggest an easier test..."

    Well, I don't see your suggestion.

    Please tell me it isn't the "if it..it is" quote.

    Will

    Edited 3 times, last by WillB (April 1, 2018 at 5:45 PM).

  • Part of being a professional record producer is making choices.
    Does this part sound better or that one?
    Is that vocal good enough?

    And what format sounds better to YOU?

    I don’t have any problem at all with anyone who prefers 44.1k or who cannot hear a difference. I DO have a problem with being told it “doesnt Matter” or even more so with the nonsense that “NO ONE can tell”

    Every professionals DAW is capable of working at at least 96k.
    The Kemper really should as well IF they expect professional use over digital connections.

    If not, or if there is a technical reason it’s only working at 48k internally, then I’m fine using it over external analogue conversion

    But it’s simply not true to say that sample rate “doesn’t matter”

  • by the way, something like that “vintage” guitar isn’t the clear test.

    Hit a cowbell or other high transient instrument in a live room and compare that.

    The combination of fast hard transient and ambient trail is easy to hear once you’re shown what to listen for.

    And three is a clear difference at different rates and also between the quality of different converters.
    Some are noticeably smeary. Some lose the reverb trail.
    The trick is to compare to the live desk out, before any conversion, and then the difference is apparent.
    But again that’s IF you are trained what to listen for.

  • Quote from WillB


    Because in order to accurately reproduce up in that range the circuit bandwidth must exceed 20kHZ significantly. The engineer/designer words, not mine.

    I understand the Engineer said this, but I don't understand why that is these days with better circuit designs and much better D-A chips than ever before? Why does the bandwidth need to be 10x higher (and lower) just to be accurate now days?

    If you use FRFR the benefit of a merged profile is that the cabinet is totally separated in the profile.

    For my edification only... ;) Kemper/Axe-FX III/ Quad Cortex user

  • WWittman:
    I'm using the KPA at 48k as master clock. I thought I saw 88.2 & 96k as an option already.

    I've thought about making a statement such as "if you can't hear the difference between 44.1k and XXk...then you..."
    But I haven't said that. I've tried to express my hearing results diplomatically.

    If KPA supports higher rates there really is no need to continue to defend 44.1 as the - all that is necessary - bullet point.

    Spikey:
    The engineer was referring to bandwidth in an analog(ue) circuit.

    Will

  • Every professionals DAW is capable of working at at least 96k.
    The Kemper really should as well IF they expect professional use over digital connections.

    If not, or if there is a technical reason it’s only working at 48k internally, then I’m fine using it over external analogue conversion

    https://www.dropbox.com/s/cuypfy8ci56o1ii/IMG_4624.jpg?dl=0
    FTFY 8)

  • If you find there’s a difference in the sound quality when recording then chances are you have a problem with your AD (and maybe even your DA).

    Your friend that stated you need 10x the rate for discrete audio is correct... only for modification of the discredited signal.

    Band passing must occur before decimation of the sound in order for the waveform to be accurately recreated without harmonic issues (aliasing artifacts).

    48k is less than one additional sample in 10 over 44.1k! If you hear that that you either have a very broken dither algorithm or super golden ears (tm), or maybe just confirmation bias. Even 96k is not sufficient to make things sound much better, it’s just over one additional sample per compared to 44.1k, which really isn’t enough when you need 9 additional samples to actually stop just basic aliasing artifacts.

    A waveform sampled at 44.1khz will be able to be perfectly reconstructed up to 22.05khz, and I mean perfectly. Go try it out on matlab if you want to see how effective sinc filtering is. That’s why the signal must be bandpassed to only be bellow 22.05k to avoid adding in harmonics which would confuse the reconstruction.

    The world of converters has moved on quite a bit since the pre digital 70’s, even in the 80’s digital DA for HIFI was quite popular and I remember the little bit as a revelation on quad electro stats in the late 80s, I’m not quite with Ethan Winer on this but I do believe even a dirt cheap converter outside of its analog components is going to give very good quality ADC and DAC, if nothing else because nearly all ADC and DAC units use the same two sets of chips. The biggest differences I hear these days are down to op amps and whatever these manufacturers do in the DA side, even there is pretty subtle.

    I would do a new blind test with a modern converter of how things sound at different rates yourself. I’ve tested here using an Apollo (not the new black face ones but an older not so great silver face) and find nothing of note between different data rates. I don’t have two computers, a pre with dual output and two cards to record the same signal at both rates simultaneously but there really was no difference that I could tell.

    I suspect more likely confirmation bias is involved.

    With regards non recorded signals e.g. synths or fx applies to signals ITB when raising the rate up to 96k will help a little if they don’t supersample, but it won’t eliminate problems. Better to use modern plugins that do supersample. Which is practically all of them. If you use a synth which doesn’t then go to the manufacturer and ask them why, they really should.


  • Well, my hang-up was the «high quality everything, that has to perfectly set up, in a nice room, with an excellent player. And the instrument must have rich overtones» part. I much prefer the idea of «hit a cowbell. See how that works». Except the part that says «you have to be trained to know what to listen for...» :D

    My point was that if the difference is sooo gentle and fragile that you have to carefully select the entire recording chain just to spot the difference, then how does that apply to real world situations where things are a little unruly and chaotic and in the spur if the moment?

    I respect the choice of working with higher sampling rates if that befits the rest of the workflow. Like with video. Or if someone flat out says «because I like higher sampling rates better.» Fine, good for you and choices in workflowes are awesome! :thumbup: But the downplay of anything else because «I’ve done testing and know what to listen for» kind of rubs me the wrong way.

    (And please don’t think I’m mad or attack you or anything, I highly appreciate the debate. :thumbup: )

  • Per:
    Thanks for your detailed explanation above.

    If "bias" is removed from your equation proven through "blind" comparisons can you speculate further as to what people who hear using higher than 44.1k results in addition clarity?

    Will

  • Blind comparisons amongst professional, trained, listeners have shown consistent ability to tell the difference between sample rates on a given converter.

    Implementation counts.

    I’ve yet to hear an available A-D converter that doesn’t sound better at 96k
    And the same for most plug ins sounding better at 96k

    Again, I’ve made my point and my feelings known.
    If you feel differently then record the way it sounds right to you

    But if you tell me I’m “wrong” then you’d better have a better track record and discography before I take you seriously.

    Every hobbyist has an opinion.
    It’s different when your livelihood depends on it.

  • Per:
    Thanks for your detailed explanation above.

    If "bias" is removed from your equation proven through "blind" comparisons can you speculate further as to what people who hear using higher than 44.1k results in addition clarity?

    I'm saying you wont with modern equipment. I speculated already, however other factors may be that you didn't use a good quality splitter and ensure that levels are absolutely identical to record your guitar in a single take at both sample frequencies on two computers with identical convertors running no FX at all i.e. EQ/Verb/Compression/etc, or that you used separate takes to record the two tracks making it easy to identify A and B.

    If you want to really test it out run the session at 192, record, then export a full 192 and a 44.1 as a raw format such as wav. Play back both and compare them. Your convertors are out of the equation then because the DA convertors run at a constant very high rate regardless of the input sample rate you give them.

    See if you can still identify which is which track with blind AB there.

    If you year no difference then you know the issue is likely with your AD convertors. Some older ones have quite high noise floor, depending on the cause of the noise running at a higher rate can make the median frequency of the noise change to be less prominent. What conversion are you currently using?