Audio Engineering Society Standards Committee

Comments on personal computer audio measurements

[Last printing 1 October 2000] Comments to date on Draft AES-6id-xxxx DRAFT AES Information document for digital audio - Personal computer audio quality measurements published 1999-09-14 for comment. The comment period is closed and the document is published.


Woodgate 1999-09-15

From: SMTP%"jmw@jmwa.demon.co.uk" 15-SEP-1999
Subj: Detailed comments on Draft AES-6id-xxxx.AES-x44

Clause 2: The reference to IEC60651 is inappropriate. A reference to IEC60268-1 would be more relevant, since that includes information on both the A-weighting filter and the ITU-R 468 method of measurement.

Clause 3.3 NOTE: This should not be a Note but main text, since it is a provision of the document. The last few words are confusing, and the concept is flawed anyway. It is not impossible that:

- at no level of analogue signal does the THD+N reach -40 dB. The THD may exceed 1% at any level, or may reach 1% before the S/N reaches -40 dB.

- the analogue circuits in the A-D may introduce extra distortion and/or noise, leading again to the impossibility of reaching a true THD+N of - 40 dB at any input signal level.

- the analogue input signal is normally from a test set, in the context of this document, so THD+N at the input is likely to be negligible.

I propose to delete consideration of the THD+N in this context. It shows up in other measurements, where it does not cause the above difficulties.

Clause 3.6: Unless this text is inadvertently faulty, it represents intellectual dishonesty of a high order. It might be the industry marketing practice in consumer products, but then I would expect a method of measurement of PMPO to be in this document!

It is quite inadmissible to relate the noise and distortion products measured at any level, other than FS, to FS. 100% noise and distortion in a signal at -40 dB would be reported as '-40 dB FS'. The definition is also, of course, contrary to the IEV and IEC60268-2. Furthermore, the undefined term 'r.m.s. full-scale signal level' is introduced. If, and only if, the *option* offered by NOTE 2, to state the test signal 'amplitude' (not the r.m.s. value!), is exercised, an initiate can calculate the real THD+N ratio. This sort of fudge is just not acceptable in a professional context.

The text should read: '....noise, measured at a specified input or output level, to that input or output level, expressed in decibels.

Clause 3.9: The relationship between the analogue voltage across the loudspeaker and digital FS depends on the setting of the playback gain control.

IEC60268-3 and IEC61606 contain exhaustive texts on this subject. There is no need to re-invent any wheels, especially if they turn out to be triangular when invented.

Clause 3.9 NOTE: This appears to be complete nonsense, presumably because it does not say what it was meant to. It would only make a little sense (a very little) if there were no analogue gain control preceding the loudspeaker or headphones.

Clause 4 NOTE: Does the AES really want to overtly antagonize the IEC in this way?

Clause 4.1: The rates of cut-off of the frequency response below 20 Hz and above 20 kHz should be specified. The specification in IEC60268-1 and ITU-R 468 for unweighted, band-limited measurements is likely to be suitable.

Clause 4.1 It is generally *not* possible to correlate A-weighted and ITU- R 468 noise values. The difference can be between 3 dB and at least 14 dB.

Clause 4.1.1: The first paragraph is in need of improvement. The full- scale value is +/-32767 whatever the THD+N is. In the digital domain, there is no distinction between signal and non-signal. Analogue measurements cannot be referred to 'this value', because 'this value' is a pair of numbers. It doesn't mean what it says. In the last sentence, a 'specification' cannot be measured in any sense that is meaningful here.

I suggest that the definition of 'digital full-scale' is taken from AES17 or IEC61606. Then, the text should continue: 'Full-scale input voltage is defined in 3.3 in terms of digital full-scale. If a different value of digital full-scale is used in a system, this shall be clearly stated. Full-scale input voltage may be determined for line and for microphone inputs.

The following text is also not optimum. The following is proposed: 1. Set up the general conditions for test as stated in 4.1.1.

2. Apply a sinusoidal signal at the STTF to the input to be measured, while monitoring the output of the A-D converter. Increase the input signal voltage until the output reaches digital full-scale.

[I can't see how the THD+N is relevant here. The test signal is surely beyond reproach?]

3. Note the input signal voltage as the required result.

NOTE 1: This measurement may be made in real time.....

[The original NOTE 1 just repeats text at the beginning of the paragraph.]

NOTE 2: In some cases, the analogue mixer (A-A)....

However, this NOTE is out of place and confusing here. It would be better as part of a new clause 'Input voltage for analogue mixer', based on relevant text in IEC60268-3. There is no 'FS' concept applicable to a purely analogue chain.

Figure 4: The frequency scale is not properly labelled.

Clause 4.1.4.3: This includes an unnecessary re-determination of full- scale input voltage. Step 5 is dishonest and cannot stand. It references the distortion at -3 dB FS to ) dB FS, there by falsifying the figure by 3 dB.

Clause 4.1.4.4:

I wonder how to carry out step 4. A method of measurement of the digital; output of the A-D is required in the document. How can clipping be identified at this point? Step 6 exhibits again the unacceptable falsification of results by relating the THD+N at -3 dB FS to 0 dB FS.

Clauses 4.1.9.1 and 4.1.9.2: Should the strange unit 'microhertz' be used, when the result is expressed, not in microhertz but in percentage?

Clause 6.3.2.1 NOTE: This employs undefined terms and abbreviations, and is incomprehensible.

Clause 6.3.2.2: There must be a reference to explain 'CS4237B line-in jack'. In addition, the following text is incomprehensible.

Clauses B.4 Paragraph 2 and B.5 paragraph 2: These are very unclear. To what do the various levels refer?

Clause B.6: All these abbreviations, etc. must be explained.

Annex C: This should be divided into clauses. In line 2, replace 'and' by 'to'. Delete all references to B and C weighting: the explanations are not correct. A-weighting is used for the measurement of **acoustic noise per se** at all sound pressure levels. [There are exceptions, such as aircraft noise, but B and C are not used for these.]

Figure C.2: The ITU-R- label should read '468'. The CCIR-RMS curve must be deleted. Its reference in AES17 was an error, and it is intended to delete it (subject to due process). CCIR-ARM and CCIR-RMS were invented for a specific and justified purpose - to prevent confusion among semi- technical buyers of consumer audio products. It has NO place in professional audio. Its perpetration would be another example of falsification of results.

If it were not deleted, the title of the Figure would be wrong. The CCIR/ITU-R registered very strong objections to the use of their acronyms in the context of 'CCIR-ARM' and 'CCIR-RMS'.

The uniform disregard of existing IEC standards on the subject is extremely disappointing. It is so extensive that it is doubtful whether the document in its present form is in accordance with AESSC policy 'not to maintain standards [in a broad sense - including ids] that are inconsistent with IEC standards'.

In my opinion, a great deal more work needs to be done on this document, apart from the matters mentioned above. Many terms and concepts are introduced informally, without clear definition, and there are technical inconsistencies and apparently meaningless sentences as well.

Regards, John Woodgate


Travis, Harris, Queen reply 1999-12-14

As requested by Chris Travis, vice-chair of SC-02-01, I am providing the working group replies to John Woodgate's comments on DRAFT AES-6id-xxxx. I have included the notes of the AESSC secretary (DQ) regarding the editorial questions. Woodgate's comments are indented. [DQ NOTE The secretariat has annotated this reply at some points.]

Where disagreements remain, the WG recommends to the SC-02 Subcommittee that the disagreements be presented in an annex to the published document.

Dr. Steven Harris, convenor of the writing group for AES-6id.

Clause 2: The reference to IEC60651 is inappropriate. A reference to IEC60268-1 would be more relevant, since that includes information on both the A-weighting filter and the ITU-R 468 method of measurement.

DQ: If Steve agrees, we can do so. However, I am concerned about referencing 468 in light of arguments of noise measurement in regard to AES17.

SH1: Delete the reference to 60651.

Clause 3.3 NOTE: This should not be a Note but main text, since it is a provision of the document. The last few words are confusing, and the concept is flawed anyway. It is not impossible that:

- at no level of analogue signal does the THD+N reach -40 dB. The THD may exceed 1% at any level, or may reach 1% before the S/N reaches -40 dB.

- the analogue circuits in the A-D may introduce extra distortion and/or noise, leading again to the impossibility of reaching a true THD+N of - 40 dB at any input signal level.

- the analogue input signal is normally from a test set, in the context of this document, so THD+N at the input is likely to be negligible.

I propose to delete consideration of the THD+N in this context. It shows up in other measurements, where it does not cause the above difficulties.

DQ: This was a major hangup on the project. I suggest that this comment may need to be published in the objections annex of the information document. Steve?

SH: The THD+N of every PC audio system that we have ever measured is always much better than -40dB, unless something is broken. The purpose of the NOTE is to allow for PC audio systems where something in the analog signal path prior to the ADC clips before the ADC output reaches digital full scale. A THD+N of -40dB, being significantly worse than the typical measured value of -80dB, was judged as being a reasonable indication of the onset of clipping. Backing off 0.5dB from this point defines a reasonable full scale reference in the situation when the analog section clips before the ADC output reaches digital full scale. I disagree with John's suggestions, and wish to leave the text as is.

Clause 3.6: Unless this text is inadvertently faulty, it represents intellectual dishonesty of a high order. It might be the industry marketing practice in consumer products, but then I would expect a method of measurement of PMPO to be in this document!

It is quite inadmissible to relate the noise and distortion products measured at any level, other than FS, to FS. 100% noise and distortion in a signal at -40 dB would be reported as '-40 dB FS'. The definition is also, of course, contrary to the IEV and IEC60268-2. Furthermore, the und only if, the *option* offered by NOTE 2, to state the test signal 'am calculate the real THD+N ratio. This sort of fudge is just not acceptable in a professional context.

The text should read: '....noise, measured at a specified input or output level, to that input or output level, expressed in decibels.

DQ: John, please tell me how this definition differs from sub-clause 8.5 of AES17. If it is only the reference to full-scale, then I agree with your wording and suspect an error was made. However, I would be in favor of a definition simply referencing the AES17 sub-clause, but I am not sure Steve would agree.

SH: The existing text is accurate. THD+N relative to full scale is commonly used in ADC data sheets and is the standard THD+N measurement as performed by the Audio Precision System One. See Figure 1 for an example, which shows THD+N (dB FS) versus input signal level.

IEC60268-3 and IEC61606 contain exhaustive texts on this subject. There is no need to re-invent any wheels, especially if they turn out to be triangular when invented.

Clause 3.9 NOTE: This appears to be complete nonsense, presumably because it does not say what it was meant to. It would only make a little sense (a very little) if there were no analogue gain control preceding the loudspeaker or headphones.

DQ: I do not see where those texts are being repeated. The definition only narrows itself to a frequency and a level. I had questioned if the note was really necessary, but Steve felt it would be helpful.

SH: Change the NOTE to: "This specification is required when the line output jack is used to drive headphones and loudspeakers directly."

Clause 4.1: The rates of cut-off of the frequency response below 20 Hz and above 20 kHz should be specified.

The specification in IEC60268-1 and ITU-R 468 for unweighted, band-limited measurements is likely to be suitable.

DQ: I suspect the additional complication will not be wanted but you should provide a wording in a comment on the call. I believe we can add several informative reference to the two documents in the final standard.

SH: OK to add informative references

The term 'performance specifications', in line 1 of paragraph 2, is not correct. This document is only about measurements, so the text should read: 'These measurements shall be made...'. In addition, 'performance specification' should be deleted from the last line of 4.1.

DQ: I agree, Steve?

SH: OK

Clause 4.1.4.3: This includes an unnecessary re-determination of full scale input voltage.

DQ: This objection begs the question of whether one can assume necessary steps of a procedure are taken at every point in the procedure. I agree it appears redundant, but evidently the writers might have been concerned about drift. If the concern is only assurance that the steps were performed, we should say so.

SH: OK as is.

Step 5 is dishonest and cannot stand. It references the distortion at -3 dB FS to ) dB FS, there by falsifying the figure by 3 dB.

DQ Steve?

SH: No, the figure is not falsified by 3dB. Using -3dB FS as the signal level makes sure than nothing is clipping.

Clause 4.1.4.4:

I wonder how to carry out step 4. A method of measurement of the digital; output of the A-D is required in the document. How can clipping be identified at this point?

DQ: Agreed. Provide alternate wording.

SH: We have test software that monitors the output of the ADC. Text is OK as is.

Step 6 exhibits again the unacceptable falsification of results by relating the THD+N at -3 dB FS to 0 dB FS.

DQ: Steve?

SH: Text is OK as is.

Clauses 4.1.9.1 and 4.1.9.2: Should the strange unit 'microhertz' be used, when the result is expressed, not in microhertz but in percentage?

DQ: I assumed this is an industry calibration practice, perhaps associated with Audio Precision equipment. Steve?

SH: OK as is. The meaning of microhertz is clear.

Clause 6.3.2.1 NOTE: This employs undefined terms and abbreviations, and is incomprehensible.

SH: Change text to: "EXAMPLE An example mixer path, using the CS4237B line-in jack to output jack, would be enabled by setting registers I2 and I3 in the WSS space to 6816 , which would unmute the path to the output mixer and set the volume to 0 dB."

Clause 6.3.2.2: There must be a reference to explain 'CS4237B line-in jack'. In addition, the following text is incomprehensible.

DQ: Ibid

SH: Change text to: "EXAMPLE An example mixer path, using the CS4237B line-in jack to output jack, would be enabled by setting, for recording:

registers I2, I3 =3D 6816 (unmute path to analog-to-digital converter and set volume to 0 dB)

for playback:

registers I6, I7 =3D 0016 (unmute digital playback and set volume to 0 dB) registers X14, X15 =3D 0016 (unmute digital mixer and set volume to 0 dB)"

Delete "then run full-duplex software."

In addition, a reference to the CS4237 data sheet as an informative reference would be good. This would explain the registers.

SH: Add to Annex C: "CS4237 data sheet, Cirrus Logic Inc, www.cirrus.com"

Clause B.6: All these abbreviations, etc. must be explained.

DQ: I hope we can defer this large task to the final document.

SH: The intended audience for the original document was people who are PC experts, but know nothing about audio. Hence they would understand ISA, PCI, USB, AC97, IEEE1394. For a more general audience, we can add a glossary.

[DQ: The abbreviation will need to be expanded in the text on first mention. The intended audience of AES standards must be the general technical public.]

Annex C: This should be divided into clauses. In line 2, replace 'and' by 'to'. Delete all references to B and C weighting: the explanations are not correct. A-weighting is used for the measurement of **acoustic noise per se** at all sound pressure levels. [There are exceptions, such as aircraft noise, but B and C are not used for these.]

Figure C.2: The ITU-R- label should read '468'. The CCIR-RMS curve must be deleted. Its reference in AES17 was an error, and it is intended to delete it (subject to due process). CCIR-ARM and CCIR-RMS were invented for a specific and justified purpose - to prevent confusion among semi- technical buyers of consumer audio products. It has NO place in professional audio. Its perpetration would be another example of falsification of results.

If it were not deleted, the title of the Figure would be wrong. The CCIR/ITU-R registered very strong objections to the use of their acronyms in the context of 'CCIR-ARM' and 'CCIR-RMS'.

DQ: In my opinion this annex does have many problems, but I think we need to see the degree of public objection. The writers feel that the users of the document need an explanation of weighting and feel this is a good one.

SH: I agree with DQ


Comments of Skirrow, 2000-01-02


>From:  SMTP%"lindos@zetnet.co.uk"  2-JAN-2000 14:02:08.64
>To: standards@aes.org
>Subj: Comment on draft AES-6id / Computer Audio

Dear Sir

PERSONAL COMPUTER AUDIO QUALITY MEASUREMENTS

I would like to comment on the Draft AES Standard AES-6id as follows:

1. This standard adds little to the already existing standard IEC268 (BS6840) which specifically includes in it's scope "sound systems for professional and household applications". By attempting to cover all equipment, using a set of measurement techniques that were based on subjective effect and thus largely independant of the type of equipment under test, IEC268 took us a step nearer to at last being able to genuinely compare equipment, regardless of type, on the basis of specifications that were a good guide to subjective quality.

The proposed AES standard, I suggest, is entirely unnecessary, and will serve only to confuse matters by setting up an 'American' alternative to the already excellent European and World standards. A better course of action would surely be cooperation with the IEC, which does of course aim to create world standards, over any items that are thought to need special revision or clarification. The AES could then refer to IEC268 just as British and other standards now do, and the IEC is to be commended in my opinion for 'harmonising' the best of European standards, with particular emphasis on the excellent work of the BBC throughout the last century on Peak Programme meters, noise measurement, and standardised methods of measurements with worst-case values laid down.

2. The recommendation that 'A' weighting be used in noise measurements is to be deplored. 'A' weighting was never intended for noise measurement, and much work went into exploring why it was unsuitable as long as 35 years ago, when 'CCIR-468' weighting was devised. 'A' weighting fails to represent the subjective effect of noise for reasons that are rarely explained.

Essentially, the ear does not respond to rms values, since of course it carries out a form of spectral analysis, and the hair cells of the cochlea signal quasi-peak displacement, not power. It is the 'concentration' of noise in each spectral region, weighted suitably, and peak-integrated over a short period, that we hear as noise, and the CCIR curve with its steeper rolloffs and 12dB peak at 6.3kHz reflects the narrower bands that the ear uses at LF which greatly reduce low frequency sensitivity to broadband noise as compared with tones (for which 'A' weighting was derived).

CCIR weighting (now more correctly ITU-R-468) has been universally used for noise measurement in broadcasting and professional audio for decades in the UK and throughout europe. It became even more widely used with the introduction of Dolby noise reduction in the sixties, since it gave representative results where 'A' weighting failed badly. While the cost of measuring equipment has held back universal adoption, that is no longer a problem (Lindos Developments will soon launch a low cost handheld test set incorporating two-channel CCIR-weighting). It would be a great pity to take a backward step now in the one area that is set to explode in importance - furthering Computers and the Internet.

3. References to the use of a 'digital FS sinewave' make no mention of the complex issue of digital dither and noise-shaping. A true sampled sinewave gives severe distortion, and I suggest that triangular PDF digital dither should be specified. This cannot be at FS since it will clip on the dither, so a level below FS must be used. Distortion at FS is in any case not very important, and +8 dB AL (-10 FS) is much more relevant, being maximum permitted broadcast PPM level.

4. Tests for 'grounding' are not relevant. There is no test for 'grounding', there is just noise, and a properly weighted noise reading is all that is required.

5. THD + noise is a measurement that has long been known to be unrepresentative of subjective effects, and is pretty useless as a measure of crossover distortion and digital distortion, where the high order of the transfer function generates spikey products that fall in the sensitive 6.3kHz region of hearing. IEC268 sensibly adopted CCIR-468 measurement of distortion products, which gives much more meaningful (and far worse) figures, and I urge that this method be recommended universally.

5. The draft specification, while doing a worse job of defining measurement methods than IEC268, fails to address the real issue. It is well known that computer audio cards, even more than TV's, videos, and other consumer items, have widely varying input levels, and never quote input level in any useful way. Aligning FS levels is wrong, and leads to clipping or noise problems. Only the use of a reference level with 'headroom' gives true interpluggability between ALL equipment and EBU standards have finally defined 'Alignment Level', a term carefully chosen to avoid ambiguity, as -18dB FS for most purposes, and -12dB FS for some low-end applications. The industry desparately needs standard levels, not just methods of measuring them, if audio is to be universally interfaced as video is, and I suggest that -18dB FS (and possibly -12dB FS for special cases) be recommended as Alignment Level for all digital systems, and that it should correspond to 0dBu (bipolar) and -6dBu (unipolar) analogue levels.

I am independantly proposing a 'LinDev Unison' standard, for use in all professional and consumer equipment, that simply defines not just measurement methods, but target values and alignment levels. With a novel input circuit this provides full compatibility between the two types of equipment without gain adjustment, using 0dBu as alignment level for bipolar (centre grounded) outputs, and -6dBu for single-ended outputs. I am in discussions with other parties over this, and I offer it for adoption by the AES, and would be pleased to cooperate in the furthering of a truly valuable new standard.

Full details of the 'UniSon' interface, and standard will be available on a website shortly. This is not intended to replace existing XLR systems in studio environments where ruggedness counts, but it could solve many problems in areas like post-production, and miniature camcorders, while becoming the de-facto standard on computers, where its compact two-channel connector and bipolar (centre grounded) interface could eliminate level-mismatch and grounding problems once and for all at negligible extra cost.

Pete Skirrow >!Pete Skirrow BSc Hons MAES AMIEE Lindos Developments Woodbridge Suffolk


Comments of Krueger, 2000-01-05


From:   SMTP%"arnyk@flash.net"  5-JAN-2000 05:55:34.56
To:
Subj: Comments on DRAFT AES-6id-xxxx, DRAFT "AES Information document for digital audio - Personal computer audio quality measurements":

Comments on DRAFT AES-6id-xxxx, DRAFT "AES Information document for digital audio - Personal computer audio quality measurements":

3.3-3.4 Setting levels is fundamental to this (and any) measurement process. Therefore sole reliance on a level-setting standard that can't be implemented with a significant population of personal computers does not strike me as a good idea. The problem I see is that there are a number of PC sound cards that have >= -40 dB THD +N at all levels. Therefore, total reliance on a level-setting technique that demands THD+N be <= 40 dB at some level seems inadvisable.

While I know of a number of PC sound card chipsets that have >-40 dB THD+N at all input or output levels (CMI 8330, 8338, HT 1869, ESS 1948F, ESS 1968S for example) I know of none that have >= -20 dB THD+N at all input or output levels. I therefore recommend that the following alternative procedure be recommended in 3.3:

"If the THD+N is worse than -40 dB, then full scale is defined as the input signal level that is 2.2 dB less than the level that induces -20 dB THD+N in the input signal".

In my experience, 2.2 dB is approximately the difference between peak level and clipping level for a pure sine wave that is distorted by clean clipping so it has -20 dB THD+N .

I also recommend that the following alternative procedure be recommended in 3.4:

"If the THD+N is worse than -40 dB, then full scale is defined at an output signal level that is 2.2 dB less than the level that induces -20 dB THD+N in the output signal".

(aside) The clipping behavior of some computer audio equipment is pretty grim. I suspect it would be more representative of subjective experiences if the standard demanded levels be set using the -20 dB THD+N point because this point can be unexpectedly low in equipment that clips poorly.

3.3-3.7 Measurements of noise without mention of measurement bandwidth seems fallacious to me. I think that a measurement bandwidth and the shape of the band in which THD+N are to be measured seems appropriate. I'd suggest 20-20 kHz (-3 dB) and -6 dB/octave slopes outside that be stipulated where ever appropriate. This could be presented as a "weighting curve" in the appropriate appendix. If I've somehow missed this in the proposed standard, please accept my apology.

I understand this is not a moot point because some standards use "brick wall" definitions of a given bandwidth, which I believe leaves the door open for measurements that imply good performance from what I think is technically deficient equipment.

3.7 It strikes me that it would not be premature to reference the -60 dB level mentioned in 4.1.5 at this point.

3.8-entire document. I have no problems with mentioning A-weighting or how it is mentioned at this time.

However, I see a serious omission. Jitter is a common technical failing in audio. Jitter probably would benefit from weighted measurements. Inclusion of a weighting methodology for jitter measurement seems appropriate, and if so, it should also be mentioned in the "Definitions" section.

4.1 I see no mention of jitter. I see no mention of intermodulation distortion. I could get "proper" and point out that more modern and enlightened terminology for several of these items would be under the headings of "nonlinear distortion" and "linear distortion". Since digital audio involves band-limited systems, it is seems to me to be inappropriate to limit the discussion of nonlinear distortion to just THD. If I've somehow missed this in the proposed standard, please accept my apology.

I recommend that jitter be measured using an 11 kHz tone, and presented as a spectrum analysis of recovered audio. and a weighted number.

I recommend that two-tone tests involving tones at 9 & 10 kHz, 14 & 15 kHz, and 19 & 20 kHz be added with the condition that the test at the highest frequency that produces <= 1%, or if <= 1% is not possible, <= 10% or IM be reported. This is required because some personal computer audio systems are incapable of response to 15 or 19 kHz. Nevertheless, some indication of high frequency nonlinear distortion seems advisable.

As a standard for "Personal computer audio quality measurements" this standard seems to ignore the more esoteric kinds of spurious responses and distortion that come with perceptual coding, which is clearly part of Personal Computer audio at this time, and can reasonably expected to continue indefinitely.

Writing a standard for evaluating perceptual coders could be a time-consuming area, and it strikes me that it would be fair to make minimal inclusions of tests that address a few perceptual coder issues. IMO, This standard should make a formal statement that such a standard needs to exist or exists, and applies to computers that provide these kinds of complex audio processing.

Of all the measurements I've done on perceptual coders I've found that those based on application of a multitone to one channel and the same multitone attenuated by 10 and 20 dB to the other channel are among those that seemed to most closely match listener comments based on music.

4.1.2 I see no mention of operation in the "600 ohm" professional environment, which is clearly part of "Computer Audio".

4.1.3.1-5 I see no mention of the use of multitones for this critical measurement. I am very impressed by the work of Richard Cabot, David Clark and others in terms of the use of multitones for measuring the technical properties of audio equipment and find its omission disturbing. I understand that some multitone tests may be protected by patents, but surely there must be some way to suggest that measurements involving them be acceptable under this standard.

Please reference my second comment on 4.1 for a context in which I find this becomes a critical (and not just significant) issue.

4.1.3.1-5 I find the use of a -20 dB level to be highly questionable. It seems way too low for the one and only set of measurements of frequency response. I have found that measurements at -10 dB can be informative, and the presence of variations in response measurements made at various levels can also be informative.

Please reference my second comment on 4.1 for a context in which this becomes an critical (and not just significant) issue.

4.1 3.4, 4.1.3.5, 4.1.5.3 etc., Let me express my personal appreciation to the standards committee for identifying "record and playback" as being an important operating mode that needs to be evaluated. In the recent past, I've taken a lot of public abuse for publishing measurements made under these circumstances.

4.1.7 Please see my comments on 4.1.3.1-5

Annex C - A weighting curve for jitter should be provided pursuant to inclusion of jitter in this standard.

Comments on remainder of standard:

I believe that were the some or all of the preceding points addressed to in the standard, there would be considerable changes to the rest of the standard, which would then become reasonably easy to determine by inductive reasoning. Therefore further discussion of the rest of the standard should properly follow the full response to these comments.

In conclusion:

For a more complete statement about my ideas about tests for computer audio performance, please see http://www.pcavtech.com/techtalk/reference/index.htm . I find in general that the proposed standard is somewhat compatible with my document. ( I don't see my document as ever being a "Standard" - it is just a list of "what I do"). I would of course try to bring my document and testing procedures into greater if not complete conformance with the AES standard, once the AES standard were approved.

My thanks to the standards committee for the work that they have done so far!


Comments of Dunn, 2000-01-05


From:   SMTP%"JDunn@iee.org"  5-JAN-2000 05:40:58.61
To: AES Standards Manager
Subj: AES6-id call for comment

In the draft AES-6id and in AES17 there are two incompatible applications of the meaning of dB FS. In particular these show up when using the term to describe analog domain levels.

1. The analog signal level at the input to an ADC or output of a DAC and scaled to be digital-domain referenced. For example an analog signal level of -60 dB FS applied to the input of an ADC would be at the analog level that corresponds to a digital level 60 dB below digital full scale. This is analogous to making input-referred noise measurements in an amplifier and is compatible across both the digital and analog domains.

2. The analog signal level scaled with respect to a clipping level in the system (either in the digital or analog domain). This method is used when the digital domain signal is not available (when there is no method of determining digital signal levels) or for other reasons.

Where clipping is due to saturation of a converter at digital full scale the results are similar but they can differ by significant amounts in some applications (including those covered by AES-6id).

SC-02-01 had started looking at a solution to this but progress was slow so we decided that it should not hold up progress of AES-6id. However the wide circulation of AES-6id (if without an explanatory note) could make any future re-definition of dBFS more difficult as it may encourage practices that the WG currently wishes to revise.

The WG has seen that there is a potential problem but has not had the resources to deal with it. I suggest that a note is added at the top of clause 3 (definitions), or in some other place where it is understood to apply to the whole document, to the effect:

"NOTE: The application of the dB FS scale in this document is particular to AES-6id. Work is in progress to resolve inconsistencies between application of this term in the digital and analog domains."

Julian Dunn


For more information about standards activity: standards@aes.org

Back


(C) 1999, Audio Engineering Society, Inc.