RogerJoensson skrev:Lite OT kanske, men finns det någon trovärdig undersökning publicerad som visar att någon klarat att peka ut en nivå eller klangskillnad på 0,1 dB vid/runt 20KHz?
AES UK Conference - the measure of Audio.
Artikeln heter:
ARE WE MEASURING THE RIGHT THINGS? ARTEFACT AUDIBILITY VERSUS MEASUREMENT
Författad av Paul Frindle från Sony Broadcast & Professional Europe, Oxford, U.K. och Publicerat 1997.
Saxat från artikeln.
Paul Frindle i AES skrev:The A,B,X procedure.
A double-blind A,B,X test is necessary to avoid bias. However impartial a listener may regard himself, in practice it is almost impossible to believe that a piece of wire and a rack full of electronics could sound the same. In the system we have devised, the operator is presented with 3 selection buttons, the outer two being source and device under test. At the point when he is convinced that he is familiarised with a difference, the middle button is made to randomly select the same as either the input or the device under test. In this way the operator only has to identify which is the same on successive numbered passes. A computer program is consulted after the test to identify which the middle button actually was at each pass.
Although it could be argued that many tries are necessary for statistically significant results in a finely balanced test. In practice it becomes obvious very quickly if a real difference is detected. An added bonus
of this method is that an operator can conduct a test alone without fear of bias or pressure from a third party. All this helps to transform the double blind A,B,X test from an intimidating and accusational experience into one of enthusiasm and discovery.
The Importance of listening equipment.
One popular opinion which seems technically plausible but is a misconception, is that since the loudspeaker has many times more distortion than the average electronic device, there can be no validity in using listening tests to assess electronic circuits. Although it can be admitted that the design of loudspeakers, due to their intrinsic limitations is still a process involving art, taste and preference, one should not be convinced that they are not sensitive analytical devices. The truth is that many artefacts produced by electronic systems are still annoying despite being many orders of magnitude below the total error of the loudspeaker. This is because the errors produced are often of a different kind to the largely low order distortion and frequency response problems of the loudspeaker. Furthermore, there is much evidence to suggest that we are very good at familiarisation and compensation for loudspeaker anomalies, and recognition of musical instrument timbre can take place even in the face of significant signal modification, provided this is constant.
It has been found that differences in an ABX test can be reliably detected on even the most questionable monitor system with similar sensitivity to a highly accurate one. The main reason for highly accurate monitoring is to gain more insight into the probable cause of the effects, rather than just detecting a difference. We have found that the use of electrostatic monitors allows a far greater possibility of identifying the nature of a technical problem, thus aiding subsequent analysis. The system used, in many cases allows direct identification by an experienced listener. Subsequent measurement is however always used to confirm this.
How sensitive is the test set up?
The sensitivity of the above test procedure and set-up is humbling indeed and much greater understanding for audio perception and measurement criteria have resulted. The following list comprises some of the effects we can reliably detect using it. Some of these existed as artefacts in our designs that were identified and subsequently redressed.
a) Absolute and stereo differential gain anomalies of less than 0.1 dB.
b) Differential stereo delays of 1 uS.
c) Frequency response variations of 0.1 dB from 'flat' 20Hz - 20kHz.
d) Harmonic distortion components at 80 dB below signal level, even when they are more than 10 dB below noise floor.
e) Limit cycle behaviour of Delta/Sigma DAC converters at 100 dB below max. signal level.
f) Absolute phase reversal of mono signal sources.
g) Departure from flat group delay characteristics giving rise to a phase/frequency response of -45deg at 20 kHz.
h) Incomplete or non-existent dithering of digital signals.
i) Stereo image change associated with statistical correlation of stereo channels using similar pseudorandom noise sources in dither process.
j) Phase modulation caused by signal interaction with system clock generation in A/D & D/A converters.
k) Phase modulation caused by imperfect system clock rejection in A/D & D/A converters.
Listening and measurement cycle.
All the above examples were unearthed as a result of listening and subsequent informed measurement. Most importantly, the majority of these effects would not have been considered important had the listening tests not been done in the manner described.
Testing the parameters involved creative use of existing test apparatus, in many cases trying to overcome the limitations of that apparatus. In other cases, results were inferred by analysing circuit signal and extrapolation. For instance, in the case of suspected extremely low clock jitter, the control loop of the system phase locked loop was demodulated and analysed on an Audio Precision test set to determine the modulation noise spectrum and thus identify the possible sources.
Some of the more esoteric artefacts were beyond the capabilities of our measuring equipment. In these cases, the effects were confirmed by simulation using signal processing modelling and subsequent comparison. In this case a deliberate generation or worsening of a suspected artefact and subsequent listening helped to identify causes subjectively.
Since no system is perfect, it is necessary to quantify the effects in terms of audible thresholds in order to decide a working compromise for system designs. This was achieved by successive listening tests and artefact variation until the limits of tolerance were established.
As a result of this work, we now have many more specifications and action more testing in manufacture and quality assurance. But even after all this it is still necessary to listen to the system occasionally in order to prevent old problems from re-appearing, either because of design modification, or simply because a previous solution was not a real fix. It has certainly happened where an effect is apparently caused by an artefact which is subsequently fixed and confirmed, only to return later due to some other modification to the system.
There is obviously the question of to what degree we are prepared to address all the above parameters in designing an audio system. Clearly there are many more issues than just frequency response and THD+N. However, it would seem that for any design effort which is seriously setting out to achieve excellent audio performance, ignoring any off these issues would eventually provoke justified criticism from the user.
Level problems related to impedance.
The important point to remember is that no output driver is without source impedance, and thus can never be completely free from loading effects. It is therefore essential that any measurement of equipment outputs be carried out when connected to a realistic load. It is also prudent to take the test kit into the actual environment to validate performance since I have found that surprises can still result from conditions in the location that I had not calculated for. The following is a selection of effects I have encountered due to output loading that did not show up on bench tests.
a) Gain drop due to loading.
b) High frequency response anomalies due to cable capacitance.
c) Low frequency response anomalies caused when output DC blocking capacitor forms series tuned circuit with destination input transformer, especially if transformer is close to saturation.
d) Changing response of transducers due to amplifier input impedance engendered damping.
Skall man bibehålla sin seriositet i branschen så måste man vara lyhörd för vad konsumenten säger och försöka sätta sig in i konsumentens situation. Jag vill gärna få till produkter som satisfierar konsumenters behov d v s upplevda behov. Det är inte jag som bestämmer eller bemyndigar konsumentens behov. Det är konsumenten själv genom sitt eget upplevda behov. Det är min uppfattning. Det behöver inte betyda att jag inte ställer frågor och ibland ifrågasätter lite för att testa konsumenten och få konsumenten att tänka till och överväga och själv överväga hur konsumenten skall ta åt sig ny information. Konsumenten tycker oftast om erfarenhetsmässig baserad kunskap framför teoretisk kunskap som inte är erfarenhetsmässigt utprovad och belagd. Detta är precis vad du efterfrågar då du vill se en undersökning som visar på det som sägs. Jag har läst en del annat som Paul Frindle skrivit och min bedömning är att han har huvudet på skaft. Han vet vad han pratar om och han kan basera det han säger på erfarenhet och kunskap.
Med vänlig hälsning
Peter
VD Bremen Production AB + Ortho-Reality AB; Grundare av Ljudbutiken AB; Fd import av hifi; Konstruktör av LICENCE No1 D/A, Bremen No1 D/A, Forsell D/A, SMS FrameSound, Bremen 3D8 m.fl.
