Clipping internally

Forums Forums SQ Forums SQ general discussions Clipping internally

This topic contains 21 replies, has 8 voices, and was last updated by Profile photo of volounteer volounteer 3 days, 15 hours ago.

Viewing 15 posts - 1 through 15 (of 22 total)
  • Author
    Posts
  • #95349
    Profile photo of Kirky
    Kirky
    Participant

    If I clip the LR Mix channel (or any group or individual channel for that matter) after processing (so no clipping at the input stage), but then send the Main Mix to a Matrix where I lower the level, will I hear any clipping? Basically, my question is: does the internal bit depth allow for channels to clip as long as the actual DA output does not (as is the case in any modern DAW)?

    Thanks for your help!
    Cheers,
    Roger

    #95353
    Profile photo of volounteer
    volounteer
    Participant

    @kirky

    Once clipped always clipped.

    The question is does AH use floating point which could avoid clipping even if you think you had clipped or does it used fixed point guaranteeing the clip comes along for the ride.

    The real question is why risk it? You don’t make anything louder by pushing the dBFS internally.

    #95369
    Profile photo of Kirky
    Kirky
    Participant

    @volounteer

    Correct, thanks for clarifying the question. Most major DAW’s use 32-bit older 64-bit floating-point representation for audio data. Which means even if you cross the 0dbFS point on a channel, clipping will not occur. Only if at the output stage you cross 0 dbFS, clipping will occur. Do the SQ desks work in the same way?

    In terms of why this is relevant: Occasionally I mix on a PA, where the kick does not really push through enough. So let’s say I make my Kick Drum louder and my Drum Bus is peaking. That’s not really a problem if the desk works in 32-bot floating point internally. I could lower the Main LR or the Matrix, or whatever comes after the Drumbus, and no clipping would occur. If the drumbus itself clips the audio, then lowering the LR mix will not help, as volunteer pointed out: once clipped always clipped.

    Maybe someone from A&H can help me out here? I read the manual and it says under “System Bit Depth”: “Uses XCVI core custom bit widths in algorithms, up to 96bits”, but I don’t understand exactly what that means.

    Thanks!

    #95370
    Profile photo of SteffenR
    SteffenR
    Participant

    have a look

    dLive Tech Questions

    #95381
    Profile photo of volounteer
    volounteer
    Participant

    @SteffenR
    Did not see the answer to @kirky question

    @kirky
    The tech question linked to leaves me confused.
    Sounds like it is a large fixed point not floating point so clipping could happen.
    But is so huge that it would be unlikely to make a difference.

    Would be nice if AH told us how they did things like that. Is itF FLP or fixed ?

    #95383
    Profile photo of Kirky
    Kirky
    Participant

    @volounteer

    I think SteffenR refers to this paragraph: “Bit Depth – 96bit refers to the accumulator only as you say. By creating our own FPGA implementation we are not limited to a single processing bid depth across the signal path. We can choose whatever bit depth we fancy for each operation or processing block, this gives the DSP engineers control over precision of the calculation and noise floor. So in a way all the arguments of floating points vs. fixed format are gone!”

    Which means different bit depths at different processing stages. I assume this means my question is answered with: no clipping when a single channel or bus clips internally. Correct?

    #95384
    Profile photo of volounteer
    volounteer
    Participant

    @kirky

    Either it is FLP or fixed.
    If FLP then it never clips until the D/A and it was over 0dBFS at that time.

    If it is a ginormous fixed then it may not clip.
    So if AH meant that it is 96bit depth fixed for a 24 bit signal then you have 48bits of headroom
    and a few more although the quality might be impacted a tiny bit if you have platinum ears.

    those 48 bits give what something like 200-300 dB of headroom? So not a problem.
    Hope the math is right – did it in my head:)

    #95413
    Profile photo of SteffenR
    SteffenR
    Participant

    16 bit 96dB of DR (CD audio)
    24 bit around 144dB
    96 dB around 550dB

    #95416
    Profile photo of ioTon
    ioTon
    Participant

    Hi,
    The accumulator has always a higher bit depth then the rest of the data structure.
    otherwise it won’t be able to calculate multiplications or summaries. To get a valid result which can be higher on number as the 2 start values, the accumulator is “bigger”.
    And nowhere is written what the internal bit depth is.
    550dB is definitely nonsense!
    SQ 5 can easily clip!
    When the metering in any stage go to peak – a few dB more an the clipping is audible.
    Greetings,
    j

    #95418
    Profile photo of volounteer
    volounteer
    Participant

    @ioton

    unfortunately we do not know the details of how AH does their computations
    floating point or fixed point with what sized accumulators registers or data paths

    i thought i inferred that the internal depth was 48 bits
    but i am old and my memory is not that good these days:)

    #95479
    Profile photo of Nicola A&H
    Nicola A&H
    Keymaster

    The short answer is we use fixed point arithmetic, but different bit depths at different stages i.e. more bits where you need them. So yes, you can clip the bus, but a) there is plenty of headroom b) the Peak LEDs warn you before a clip occurs, and c) the clip will only manifest as audible distortion if consistent enough to generate square waves for a number of samples. Generally speaking, the recommendation is to use a good gain structure and avoid clipping other than the occasional red light, be it on a channel or summing to a bus.
    Here’s an extract from a paper on XCVI we published in the US (attached if you’re interested):
    To get the sonic performance right and keep noise at bay, it’s very important to look at bit depth in all p arts of the processing path. Processing introduces errors, which are manifested as THD+N. Fortunately this audio distortion can be controlled with appropriate design choices since resolution is essentially proportional to bit depth. We studied bit depth as part of the XCVI research and we ended up with much wider
    bit depth where it matters. A good example of this variable bit depth is the bus summing which employs 96 bit. Remember all those debates on fixed point vs. floating point? Some DSPs offer floating point arithmetic to circumvent the issue of limited bid depths and overflow, however that comes at the expense of noise modulation. With variable bit depth, the debate becomes irrelevant. The ultimate solution is fixed point arithmetic, with more bits where you need them. Not quite solving world peace, but it’s a start.

    Attachments:
    You must be logged in to view attached files.
    #95489
    Profile photo of volounteer
    volounteer
    Participant

    @nicola

    Thanks.

    That is what I suspected. Thank you for confirming it.

    People should definitely not clip because they are trying to make it louder.
    There is plenty of headroom so avoid that red light.

    Perhaps AH should consider a few orange lights in between some yellows and a few reds to be the warning
    and red indicating a definite clip with more reds lit meaning terrible bad

    #95490
    Profile photo of MarkPAman
    MarkPAman
    Participant

    Well, this being the SQ section, you have the Chromatic Channel Metering which is adjustable, so can set it up pretty much as you want – though I think the clip indicator will always come on “early”, as it should.

    Page 45 of the current manual if you’re interested.

    #95491
    Profile photo of volounteer
    volounteer
    Participant

    @markpaman

    What I would like is a preclip set of lights showing there is a potential problem looming but also actual clip lights showing how bad it is should it happen.

    SQ may have that ability. All the mixers should have it.

    OTOH people who think louder is better probably don’t care and the rest of us add extra headroom so dont need it either.
    I got an example CD from a mixing service and the first song they had was about 30% red lines showing clipping had happened. I could not believe people would pay for mixing that bad.

    #95722
    Profile photo of Dave Meadowcroft
    Dave Meadowcroft
    Participant

    I was going to ask about this myself after seeing the occasional flicking red light on my sends/mix layer this weekend.
    Towards the very end it was quite a hot mix to be honest and not the normal, it was a mix I inherited part way through. Most instrument channels peaking at around +6 (-12 dBFS) with faders at unity, the vocals post compression peaking about 3dB higher and faders around +5 so I would estimate around -4 dBFS. Post fader sends to fx, and of course main, were starting to warn me of possible impending clipping which I believe happens at -3 dBFS so it was probably hitting that occasionaly.
    I was surprised though as the Aux masters and Main fader were below unity so I expected to have the headroom available as I wouldn’t be clipping the D/A.
    Having read the document that Nicola attached it makes sense, but means to avoid internal clipping I need to run gains lower (IEM users are gonna love that change during a performance) or lower all channel faders and bring up post fade send masters and main fader. A bit of a change in working from any other digital desk but manageable now I know.
    I thought clipping (after A/D and so long as we don’t overdrive the D/A beyond 0 dBFS/+22 dBu) was a thing of the past in the digital age!
    Not a complaint, and I take on board the reasoning – just an unexpected surprise.

Viewing 15 posts - 1 through 15 (of 22 total)

You must be logged in to reply to this topic.