Loading ...
Sorry, an error occurred while loading the content.

Re: Re HMs post

Expand Messages
  • tonycdk
    There is no doubt that high frequency resolution in the upper frequencies is not necessary, but in addressing only high frequencies when talking about the
    Message 1 of 8 , Jun 3, 2012
    • 0 Attachment
      There is no doubt that high frequency resolution in the upper frequencies is not necessary, but in addressing only high frequencies when talking about the number of taps totally ignores the issue that although high resolution is not required in the high frequencies IT IS required in the lower frequencies.

      Really it is quite simple, at 16 kHz a Q of 16 has a full peak width at -3 dB of 1 kHz and so, if you want say 6 data points across the peak to fully describe it, you can do so with a frequency resolution of about 167 Hz. For a 16 Hz tone you will need a resolution of 0.167 Hz. Forget the specific numbers but it is clear what requirement should drive the number of taps - it is the low frequency requirement!

      There is no problem with using high resolution in the higher frequencies so long as one does not try to correct at that resolution but uses some thought.

      It is worth once again refering to the figure that I referenced in the referenced previous note. This is the last figure in my "Photos" folder. It shows that one can correct,measure a response, and think that all is well. However, the plot shows what you get if you process the SAME impulse response with different frequency resolution (i.e. a different number of taps). With more data samples being analyzed one is analyzing a longer period of the impulse (so one would expect some differences)but the results are clear, what seems 'perfect' at one analysis resolution can show significant problems at another. I should point out that the plot at 0.092 Hz resolution is closest to what one hears during the log sweep.

      This leads me to a general complaint about measurements that people present. There is almost never enough information to fully assess the results. At the very least one needs to know information regarding smoothing and the number of data points used in any FFT analysis (or the frequency resolution - same thing really).

      One may or may not hear some items when listening to music it depends upon several parameters. However if one is prepared to address being able to hear 0.1 dB differences with a pink noise source (which is probably inaudible when listening to music) one can't complain about raising issues in the low bass (which also may be inaudible when listening to music).


      Tony






      --- In regsaudioforum@yahoogroups.com, "Robert" <regtas43@...> wrote:
      >
      >
      > I was really talking about the higher frequencies primarily
      > (in the example in particular)

      > About the bass, Tony did indeed talk about this.
      > I could be wrong in particular cases, but
      > I think the chances of a 1 Hz resonance with'
      > audible consequences are very low. This would have to be
      > an extremely high Q resonance , and in any kind
      > of well done room (as I would consider a well done room)
      > such things should not happen.
      > This is the whole point of flexible walls,
      > that they spread out the Q of room modes.
      > More on this later when Newport is over.
      > REG
      >
      >
      > --- In regsaudioforum@yahoogroups.com, "tonycdk" <tcdk@> wrote:
      > >
      > > Robert and I have gone around on this issue a few times in the past (see an example from a couple of years ago in message # 33983). I consider a large number of "taps" (this implies finer frequency resolution)to be necessary in order to be able to handle issues in the low bass frequencies. In the low bass resonances can be under 1Hz in width and so one needs excellent frequency resolution to be able to even know that they are there (except that they are very audible when sweeping through the appropriate frequencies). If you can't see the resonance in your measurement then you can't correct it! One needs at least three or four measurement points across the resonant peak to properly characterise it and correct for it, consequently one needs frequency resolution of the order of 0.2 to 0.1 Hz. You don't need this in higher frequencies but, if the full frequency range is covered in the single set of taps, this requires a very large number of taps.
      > >
      > > Tony
      > >
      > > --- In regsaudioforum@yahoogroups.com, "Robert" <regtas43@> wrote:
      > > >
      > > > We keep going around in circles on this, but as far
      > > > as I am aware there is nothing at all to suggest that
      > > > having a lot of "taps" is needed. Only spatially stable
      > > > corrections are interesting and these are not going to involve
      > > > micro fine resolution. In fact, there is a lot of evidence
      > > > that one is better off not doing that. Speakers tend to be quite smooth in spatially averaging. Not totally so. But look
      > > > for example at what could be done with FOUR parametric
      > > > EQ filters on the 2k to 9k response of the M40.
      > > > http://www.regonaudio.com/Digital%20Correction%20for%20Audio%20Part%20III.html
      > > > Figure 1.
      > > > This is flat enough. It is flat within the tolerance of measurement in the sense that if you moved the mike a little it would
      > > > probably change more than the errors shown.
      > > > Doing more processing than this in this frequency range is not
      > > > only futile--it is counter-productive. It will make things
      > > > sound worse most likely.
      > > > And that is with four coefficients.
      > > > Going crazy with taps is not needed. And it is not even a good idea.
      > > > REG
      > > > --- In regsaudioforum@yahoogroups.com, "mm" <yipmangmeng@> wrote:
      > > > >
      > > > > Uli,
      > > > >
      > > > > What you are saying is that a computer is a much better tool to put in the software like Acourate than using a single box?
      > > > >
      > > > > Yip
      > > > >
      > > > > --- In regsaudioforum@yahoogroups.com, Uli Brueggemann <uli.brueggemann@> wrote:
      > > > > >
      > > > > > Some comments:
      > > > > >
      > > > > > 1. the inverse of a minimumphase pulse is a minimumphase pulse. As already
      > > > > > mentioned the convolution of both results in a Dirac pulse.
      > > > > > - So some systems are marketed by telling the customers about phase
      > > > > > correction despite the correction is just minimumpase.
      > > > > >
      > > > > > 2. A real pulse can be decomposed in a minimumphase pulse and an
      > > > > > excessphase pulse. If a given pulse is minimumphase then the excessphase
      > > > > > pulse is a Dirac pulse.
      > > > > > - But I have never seen a room measurement up to now where the decomposed
      > > > > > excessphase pulse is a Dirac. Acourate has functions to easily apply the
      > > > > > decomposition and to see the results in the charts.
      > > > > >
      > > > > > 3. Even a simple rectangle window on a given pulse is a time window. Thus
      > > > > > the term 'time window' means nothing.
      > > > > > - Fixed size windows give wrong informations. Too much high frequency
      > > > > > oscillations and unsatisfying information for low frequencies. The
      > > > > > solution is to use frequency dependent windows. Googling for publications
      > > > > > about FDW does not show up many hits. So I wonder if SigTech has really
      > > > > > applied such windows.
      > > > > >
      > > > > > 4. Today a Sharc DSP allows e.g. filters with 6144 taps (OpenDRC). Other
      > > > > > solutions like TacT or Four Audio are using DSPs with up to 2048 taps,
      > > > > > combined with multirate filters.
      > > > > > - Filters with 64k taps or 128k taps or even 256k taps are not possible
      > > > > > with DSPs, so the power of PC CPUs or GPUs (graphical procesors, e.g. CUDA
      > > > > > or OpenGL) is required.
      > > > > >
      > > > > > Uli
      > > > > >
      > > > > >
      > > > > >
      > > > > > On Thu, May 31, 2012 at 7:22 PM, Robert <regtas43@> wrote:
      > > > > >
      > > > > > > **
      > > > > > >
      > > > > > >
      > > > > > > I am replying separately to avoid huge length!
      > > > > > > First of all, it is wrong to suppose that frequency
      > > > > > > response correction does not affect time.
      > > > > > > In a minimum phase system, correction of
      > > > > > > frequency response (to flat) by minimum phase
      > > > > > > filter inversion also correct phase behavior and
      > > > > > > leads to a perfect impulse response.
      > > > > > >
      > > > > > > If one does not understand that first, there is no
      > > > > > > point in going on.
      > > > > > >
      > > > > > > Audiophiles seldom know much of anything about these
      > > > > > > matters. But they could learn.
      > > > > > >
      > > > > > > Uli determined by measurement that there is some
      > > > > > > non-minimum phase behavior in speaker/room situations.
      > > > > > > This means that one needs to do some other kind of
      > > > > > > filter correction, with "latency", holding on to the
      > > > > > > signal a bit so one could in effect see slightly into
      > > > > > > the future.
      > > > > > >
      > > > > > > Sigtech's research suggested that in decent rooms(the system
      > > > > > > was designed for pros mostly), the non minimum phase part
      > > > > > > (which can only arise from considerable energy storage by the
      > > > > > > room) was not very significant. In any particular setup,
      > > > > > > this could be right or wrong.
      > > > > > >
      > > > > > > But it really presents a complete misunderstanding of
      > > > > > > almost everything to suppose that minimum phase EQ does
      > > > > > > not affect time. It does. And if one is correcting
      > > > > > > a minimum phase system, then one ends up with complete
      > > > > > > filter inversion.
      > > > > > >
      > > > > > > This is one reason that Sigtech liked to demo with Dunlavy speakers--which
      > > > > > > were themselves minimum phase. The whole system was
      > > > > > > then in suitable room essentually minimum phase.
      > > > > > >
      > > > > > > Second point: Of course the Sigtech allowed choice of target
      > > > > > > curves. They typically showed correction to flat to enable people
      > > > > > > easily to see how well the correction works: it is easier
      > > > > > > to see deviations from a horizontal line than from a
      > > > > > > target curve that slopes.
      > > > > > >
      > > > > > > Moreover, the measurements in Sigtech are time windowed
      > > > > > > so that the top end has direct arrival measured only. There
      > > > > > > is no reason then for the top to slope down nearly as much as
      > > > > > > a long window measurement, which would be much like an RTA,
      > > > > > > would need to slope.
      > > > > > >
      > > > > > > But target curves were available for sure.
      > > > > > >
      > > > > > > This is shown here for example
      > > > > > >
      > > > > > > http://www.regonaudio.com/Digital%20Correction%20for%20Audio%20Part%20II.html
      > > > > > > figure 3b--target curve that rolls off the top(user selectable and
      > > > > > > adjustable.
      > > > > > >
      > > > > > > As to processing power, the reason the Sigtech was expensive was
      > > > > > > that it did the processing internally , not via computer. So in
      > > > > > > effect they had to build a state of the art(for the time) computer
      > > > > > > from scratch. There was a lot of processing power--but it cost money.
      > > > > > >
      > > > > > > REG
      > > > > > >
      > > > > > >
      > > > > > >
      > > > > >
      > > > >
      > > >
      > >
      >
    Your message has been successfully submitted and would be delivered to recipients shortly.