I did a little more research and a few quick looks but nothing captured yet. The interesting thing from the AP plots of the AKM parts (look at the datasheets for the AKD4396 and AKD4399) when they plot the 1KHz without the notch the noise floor is -110 dB. Plotting with the notch filter engaged the noise floor drops to -130 dB. The noise floor at -60 dB is at -140 dB. This suggests (and sort of what I see) a 10 dB shift in the part and a much higher shift (30 dB+) in the ADC in the AP. All of this needs proper verification which is very tedious. You can't trust the instruments until you know where they are no longer accurate.
I can't quickly find other plots to compare for now.
You're right to suspect noise from the AP's ADC is going to muddy the waters. Bruno mentioned recently on his blog how he cringes when he sees plots where the AP's notch hasn't been switched in.
The trick here is to compare figure 9 with figure 13 in the AKD4396 DS. From the FFT its clear all HD distortion components are -106dB or below yet from the THD+N vs level plot (0dBfs) we get a -100dB reading. I conclude from this comparison that fig 13 shows the additional noise (not HD) added between -30dBfs and 0dBfs.
As there are no notched results shown for 44k1, I'm next looking at fig 18 (96k sampling) where the mid-band grass looks to be -116dB. Looks to be an appalling high level of noise given that the FFT bin width (eyeballed the LF end step width) is probably 6Hz (16k sample size). Hence the FFT gain is 39dB giving an integrated noise floor in the 48kHz band of 77dB. Is it possible its actually this bad or have I (or they) made some error?
<edit> Duh! Of course here we must be seeing how poor the AP's ADC is. The true figure will be with the notch in place, That's fig 19 where the grass is -130dB, But that's only a 14dB improvement, to -91dB, still far from impressive.
The trick here is to compare figure 9 with figure 13 in the AKD4396 DS. From the FFT its clear all HD distortion components are -106dB or below yet from the THD+N vs level plot (0dBfs) we get a -100dB reading. I conclude from this comparison that fig 13 shows the additional noise (not HD) added between -30dBfs and 0dBfs.
As there are no notched results shown for 44k1, I'm next looking at fig 18 (96k sampling) where the mid-band grass looks to be -116dB. Looks to be an appalling high level of noise given that the FFT bin width (eyeballed the LF end step width) is probably 6Hz (16k sample size). Hence the FFT gain is 39dB giving an integrated noise floor in the 48kHz band of 77dB. Is it possible its actually this bad or have I (or they) made some error?
<edit> Duh! Of course here we must be seeing how poor the AP's ADC is. The true figure will be with the notch in place, That's fig 19 where the grass is -130dB, But that's only a 14dB improvement, to -91dB, still far from impressive.
Last edited:
So others have identified the ADC in the AP as leading to errors. I thought it was me. . .
I have a setup here with a notch to remove the fundamental but getting a relationship between noise and signal is not straightforward. Noise only has meaning in terms of bandwidth. Using an FFT the way we were discussing it can be very misleading if we don't know the effective bandwidth of the FFT.
Can you detail your math a little more so I can follow? The conversion from -130dB (6 Hz bw?) to -91 dB (48 KHz?) is significant. I follow the noise per unit bandwidth but I want to be sure we are all talking at the same level.
I'm looking at an AKD4414 on my bench and the THD+N in a 30 KHz bw is -104 dB from 0dBFS (3V in this case). The second harmonic measures at -107 dB. This is an analog measurement. How does that relate to the -91 dB you mentioned? When I check the noise floor it translates into -156 dBC/rtHz. (My FFT has the calculations built in) That works back to -113 dB (in 30 KHz not including the second harmonic). Dropping the signal to -10dBFS the THD+N becomes -103, the second harmonic is -108 dB (the other are all 10 dB lower) and the noise floor becomes -146 dBC/rtHz. at -20 (the lowest before the analyzer gets upset) the second is -108 dBC and the noise is -136 dB/rtHz. (I was using the peak hold for the noise FFT measurements so they are really peaks, not averages). The noise floor of the analyzer may also be a limitation when its at a level comparable to the source. I'll need to setup a suitable preamp to see info below 100 mV. That is a chore for tomorrow. It will give some insight into the noise modulation issue I hope.
I need to get a "ladder dac" here to measure if only to see if it gives different results.
I have a setup here with a notch to remove the fundamental but getting a relationship between noise and signal is not straightforward. Noise only has meaning in terms of bandwidth. Using an FFT the way we were discussing it can be very misleading if we don't know the effective bandwidth of the FFT.
Can you detail your math a little more so I can follow? The conversion from -130dB (6 Hz bw?) to -91 dB (48 KHz?) is significant. I follow the noise per unit bandwidth but I want to be sure we are all talking at the same level.
I'm looking at an AKD4414 on my bench and the THD+N in a 30 KHz bw is -104 dB from 0dBFS (3V in this case). The second harmonic measures at -107 dB. This is an analog measurement. How does that relate to the -91 dB you mentioned? When I check the noise floor it translates into -156 dBC/rtHz. (My FFT has the calculations built in) That works back to -113 dB (in 30 KHz not including the second harmonic). Dropping the signal to -10dBFS the THD+N becomes -103, the second harmonic is -108 dB (the other are all 10 dB lower) and the noise floor becomes -146 dBC/rtHz. at -20 (the lowest before the analyzer gets upset) the second is -108 dBC and the noise is -136 dB/rtHz. (I was using the peak hold for the noise FFT measurements so they are really peaks, not averages). The noise floor of the analyzer may also be a limitation when its at a level comparable to the source. I'll need to setup a suitable preamp to see info below 100 mV. That is a chore for tomorrow. It will give some insight into the noise modulation issue I hope.
I need to get a "ladder dac" here to measure if only to see if it gives different results.
Your point about knowing the bandwidth is apposite, particularly when reading off FFTs - many ignore the fact that the FFT's a histogram, not a graph.
So yes - I estimated the number of bins in the FFT by looking at how wide the straight lines were at the very lowest end (around 20-100Hz on the plot) where the frequency scale is most expanded. The length of the shortest straight line is under one graduation (10Hz there) so I figured its the next power of two (relative to the 96k sample rate) beneath that, 6Hz. This gives us the effective bandwidth of each of the 'bins' in the FFT. Hence -130dB off the plot is the noise in a 6Hz bandwidth at that frequency. We need to scale this up by log10(8192) which gives 39dB to get to the full bandwidth (48kHz). This ignores the effect of the windowing function employed (which is an unknown) but its decent to a first approximation.
The -91dB is from the plot of the '96 and only at 96kHz so wouldn't be apples to apples with your measurement here. It looks as though your -104dB figure is noise dominated as the 2H is 3dB lower.
I'm unclear then how -104dB THD+N can arise if your noise is -113dB and your distortion -107dB.
You might want to investigate how the peak hold is done - if its after the FFT, then the FFT contributes averaging over its record length. Choose a shorter FFT (fewer bins) to get less averaging.
That would indeed be interesting - preferably choose one which isn't R2R - weighted current source would be my choice i.e. TDA1541 or TDA1545. If your measurement gets sensitive enough you may well be able to measure noise differences between R2R and weighted CS type DACs
How about this as an idea . Record some CD's on the best recording machine one can reasonably afford . Buy ( make ) an oscillator at the Rosen level of performance as recommended at the beginning of 16 bit ( THD > - 117 dB ) . Accept the recorder will have errors and one hopes not unlike real CD's . Then simply see what nasty stuff comes out at - 40 and - 60 dB . Try a Dolby Pro Revox if available . I suspect the trends would be reliable . I suspect also if no great trend is seen the recorder is not very good .
Is it silly of me to think that feeding white noise into the DAC input is good enough as dither ? I am not wanting to do subtle stuff . Taking early digital to 12 bits would be fine . My conjecture is throw away the bad stuff . It would be on a level control . As I said to someone today even hiss at - 60 dB is better than Tamla Motown distortion . They used a Sony PCMF1 and threw away the master tapes it is said ! Perfect sound that lasts forever was believed by T M . If the error is in bit 15 and 16 we still have 84.3 dB of dynamic range ( then hiss ) . I suspect natural dithering must happen via master tape hiss so the idea might not help TM and they might be better than I think because of it . Early DDD recordings perhaps are the ones where some extra dither might work ?
Is it silly of me to think that feeding white noise into the DAC input is good enough as dither ? I am not wanting to do subtle stuff . Taking early digital to 12 bits would be fine . My conjecture is throw away the bad stuff . It would be on a level control . As I said to someone today even hiss at - 60 dB is better than Tamla Motown distortion . They used a Sony PCMF1 and threw away the master tapes it is said ! Perfect sound that lasts forever was believed by T M . If the error is in bit 15 and 16 we still have 84.3 dB of dynamic range ( then hiss ) . I suspect natural dithering must happen via master tape hiss so the idea might not help TM and they might be better than I think because of it . Early DDD recordings perhaps are the ones where some extra dither might work ?
Over the last several days, I've been looking at what was yesterday and what is today.
I couldn't help noticing that in most cases and with most well respected names in the loudspeaker industry, while most things went this or that way, one thing was a common trait - in comparison with the late 70ies and early 80ies, the heyday of audio, loudspeakers have become more efficient.
I am not talking about an odd dB here and there. I am talking about not insignificant improvements. For example, old AR speakers, like the AR3A, AR5 (which I had), did around 86-87 dB/1W/1m, while their own series some 10 years later were already at 89-91 dB, or twice as loud.
The current crop is already moving in the 90-93 dB range.
I find this a little odd, given that power has become cheap and abundant with the advent of class D amps. We now have speakers about four times as efficient as they used to be, driven by amps also about four times as efficient as they used to be.
All of which makes Wayne's case ever more exotic. I joke about it, but it really is so.
Any thoughts? Views? Opinons? Just trying to understand better.
I couldn't help noticing that in most cases and with most well respected names in the loudspeaker industry, while most things went this or that way, one thing was a common trait - in comparison with the late 70ies and early 80ies, the heyday of audio, loudspeakers have become more efficient.
I am not talking about an odd dB here and there. I am talking about not insignificant improvements. For example, old AR speakers, like the AR3A, AR5 (which I had), did around 86-87 dB/1W/1m, while their own series some 10 years later were already at 89-91 dB, or twice as loud.
The current crop is already moving in the 90-93 dB range.
I find this a little odd, given that power has become cheap and abundant with the advent of class D amps. We now have speakers about four times as efficient as they used to be, driven by amps also about four times as efficient as they used to be.
All of which makes Wayne's case ever more exotic.
I joke about it, but it really is so.Any thoughts? Views? Opinons? Just trying to understand better.
You need 10db to be twice as loud and a linsource is a different beast from point source as I had expressed earlier at 4m listening distance for eg. a 90db/w/m point source will have as much output as a 84db/w/m linesource with 1 watt and will lack the size and growth capabilities of said large linesource .
Two things people overlook . The room's ability to work as a pressure vessel comes into it , cars are better as they have door seals . Also a magnet is a magnet . I worked where the worlds most power magnet was produced , at the time 23 Tesla . Personally I doubt they did . 9 Tesla seems to me about the limit . A good transformer 1 T . This limits distortion etc in speakers .
Better to make the system excellent at low level . The ear adjusts very quickly to level . It is always is looking for the low to high comparison . As long as an 85 dB average level is reached the bass will seem correct . Trying to make 0.1 to 20 Hz high in output with correct temporal correlation . That will give power to the system . Very clean PSU and sources if you do . That would not exclude LP .
To give a low to high comparison . I was in Bonn in 1982 . I saw this most beautiful woman walking up the street . I was dress like in a Western film in my motorcycle gear . I was a bit wrecked form trying 2 hours on my Honda 900 as fast as it would go . I didn't take any trouble pretending I wasn't looking . Then I got the most appreciative return of look . Then she got closer . She was in perfect ( read curvy ) proportion with fantastic legs and the obvious ( even the not so obvious I seem to remember ) . Only problem which was only apparent when very close was she must have been 7 feet tall !!" If I had looked properly I would have noticed her head was above the windows of the shops . The modern me would have said hello , the 1982 version didn't .
Orders of magnitude (magnetic field) - Wikipedia, the free encyclopedia
Better to make the system excellent at low level . The ear adjusts very quickly to level . It is always is looking for the low to high comparison . As long as an 85 dB average level is reached the bass will seem correct . Trying to make 0.1 to 20 Hz high in output with correct temporal correlation . That will give power to the system . Very clean PSU and sources if you do . That would not exclude LP .
To give a low to high comparison . I was in Bonn in 1982 . I saw this most beautiful woman walking up the street . I was dress like in a Western film in my motorcycle gear . I was a bit wrecked form trying 2 hours on my Honda 900 as fast as it would go . I didn't take any trouble pretending I wasn't looking . Then I got the most appreciative return of look . Then she got closer . She was in perfect ( read curvy ) proportion with fantastic legs and the obvious ( even the not so obvious I seem to remember ) . Only problem which was only apparent when very close was she must have been 7 feet tall !!" If I had looked properly I would have noticed her head was above the windows of the shops . The modern me would have said hello , the 1982 version didn't .
Orders of magnitude (magnetic field) - Wikipedia, the free encyclopedia
Lowest measured distortion NRC: Magico S5
You gets what you pay for ....
SoundStageNetwork.com | SoundStage.com | NRC Measurements: Magico S5 Loudspeakers
You gets what you pay for ....
SoundStageNetwork.com | SoundStage.com | NRC Measurements: Magico S5 Loudspeakers
Because speakers are easier to make which work subjectively better, than electronics are. There's a very strong reaction to this concept, still, but people to some degree instinctively realise this - the more efficient the speakers, the less stressed are the electronics in terms of the power processing requirements, and one automatically gets better sound ...I find this a little odd, given that power has become cheap and abundant with the advent of class D amps. We now have speakers about four times as efficient as they used to be, driven by amps also about four times as efficient as they used to be.
All of which makes Wayne's case ever more exotic.
Any thoughts? Views? Opinons? Just trying to understand better.
I seem to remember being involved in a discussion of the precise terminology to be used, before, . What counts is the acoustic levels that are generated, for the levels of electrical power that are required to be fed to the speaker terminals to achieve that - however one wants to measure that ...
. What counts is the acoustic levels that are generated, for the levels of electrical power that are required to be fed to the speaker terminals to achieve that - however one wants to measure that ...
Well sensitivity has nothing to do with efficiency ....
Why not? I must be missing something here - a 93dB/W speaker (more sensitive) is also wasting less of the amp's energy as heat than one posting 87dB/W. It must therefore be a more efficient transducer no?
@nigel - about dither, its not something you can do after the event. Quantizing without dither introduces distortion, one distortion's in there you can't get it out again. DACs (other than S-D type ones) don't need dither because they're not doing quantizing.
Last edited:
Not really , one speaker could be 16 ohm nominal the other 4 ohm nominal both having same sensitivity of 90db/m/2.83v , same sensitivity not the same efficiency ....
Fast used the wrong terminology and instead of correcting chooses to moon walk across the dance floor ..
Wayne, ol' buddy, I fail to see the difference. If with the same 2.83V both produce the same SPL at 1 m, while recognizing that one will require 4 times more current than the other, in my book, they are equally efficient as loudspeakers, even if one is actually pushing the amp harder than the other.
Only same sensitivity per volt, not per watt. If you define sensitivity only in volt terms then they're correlated to some degree, a long way from the claim of 'nothing to do with...'. That was the original moon walk.
The term efficiency was deliberately used, because that is what counts - if some want to play word games revolving around the impedance seen by the amplifier that's up to them. People listen to a certain recording at an appropriate volume, and they will just adjust the volume control to suit, forcing the amplifier to work at certain power requirement levels - end of story ...
- Status
- Not open for further replies.