Bernhard,
Please note that if you just synthesize 1kHz -60dBFS 16 bit file, it will have the embedded distortion of about 0.75%. If you plan to measure the TDA1541, this distortion is close to the chip’s specs. Naturally, for the distortion measurement of the chips with better distortion figures and nowadays there are a plenty of them, such file is useless.
Another file can serve much better for this purpose. It is synthesized originally like 32 bit file and is converted down to 16 bit using dither and noise shaping. The result will depend on the FFT parameters used for the analysis, with Hanning windowing and 65536pts something like 0.012% is achieved. Below are the graphs showing the difference between these two files.
The files are here:
http://uk.geocities.com/react_71/
Pedja
Please note that if you just synthesize 1kHz -60dBFS 16 bit file, it will have the embedded distortion of about 0.75%. If you plan to measure the TDA1541, this distortion is close to the chip’s specs. Naturally, for the distortion measurement of the chips with better distortion figures and nowadays there are a plenty of them, such file is useless.
Another file can serve much better for this purpose. It is synthesized originally like 32 bit file and is converted down to 16 bit using dither and noise shaping. The result will depend on the FFT parameters used for the analysis, with Hanning windowing and 65536pts something like 0.012% is achieved. Below are the graphs showing the difference between these two files.
An externally hosted image should be here but it was not working when we last tested it.
The files are here:
http://uk.geocities.com/react_71/
Pedja
Pedja,
thanks, I will burn the CD tomorrow and maybe this helps.
Just found another test CD with better 1 kHz 0dB sine than the other one.
1541 and silvercrown are totally identical ( +/- 0,0dB ) except K3:
1541 is -69,8dB
Silvercrown is -70,2dB
Repetitive accuracy of the analyzer is 0dB.
No idea what is the embedded distortion of the sine from this CD.
thanks, I will burn the CD tomorrow and maybe this helps.
Just found another test CD with better 1 kHz 0dB sine than the other one.
1541 and silvercrown are totally identical ( +/- 0,0dB ) except K3:
1541 is -69,8dB
Silvercrown is -70,2dB
Repetitive accuracy of the analyzer is 0dB.
No idea what is the embedded distortion of the sine from this CD.
Attachments
Thijs, as the opposite of the phenomenon usually found in the analog devices, in the digital the distortion increases as the signal level decreases. The numbers mentioned above are the limits of the format we are operating with.
The dither is the technique that can improve the distortion at low level at the expense of the higher noise floor. Noise shaping is… the name is self descriptive.
For the file I uploaded it was used the high-pass triangular dithering. In this graph the red line shows the same dithered signal but the noise shaping was not used.
The graphs are analysis of the files themselves, so no D/A conversion was involved in this.
There is interesting presentation that can be found here:
http://www.angelfire.com/music4/audio_visions/know.html
(check Dither and noise shaping demo by Chris Hicks)
Bernhard, if there was not any huge mistake while the file was generated, its embedded distortion should be negligible in the results you achieved. However, the numbers should be lower and there is probably some other source of distortion that dominates there. Considering the difference between S1 and non-S1 versions of TDA151A, according to Philips, it is in the linearity at the low level.
Pedja
The dither is the technique that can improve the distortion at low level at the expense of the higher noise floor. Noise shaping is… the name is self descriptive.
For the file I uploaded it was used the high-pass triangular dithering. In this graph the red line shows the same dithered signal but the noise shaping was not used.
An externally hosted image should be here but it was not working when we last tested it.
The graphs are analysis of the files themselves, so no D/A conversion was involved in this.
There is interesting presentation that can be found here:
http://www.angelfire.com/music4/audio_visions/know.html
(check Dither and noise shaping demo by Chris Hicks)
Bernhard, if there was not any huge mistake while the file was generated, its embedded distortion should be negligible in the results you achieved. However, the numbers should be lower and there is probably some other source of distortion that dominates there. Considering the difference between S1 and non-S1 versions of TDA151A, according to Philips, it is in the linearity at the low level.
Pedja
Bernhard, I do not have better idea than that you should check your analyzer set-up.
Thijs, you could try some sound editing program like SoundForge or CoolEdit, trial versions should be available for download.
Alf, with this sampling frequency the high pass noise shaping you presented above is generally less favorable for this purpose since it shows about 50% higher total distortion number than if noise shaping is not high passed. In the case of -95dB signal shown below it is about 1% instead of 0.64% achievable otherwise (actually the second curve follows the Fletcher-Munson equal loudness contour).
Pedja
Thijs, you could try some sound editing program like SoundForge or CoolEdit, trial versions should be available for download.
Alf, with this sampling frequency the high pass noise shaping you presented above is generally less favorable for this purpose since it shows about 50% higher total distortion number than if noise shaping is not high passed. In the case of -95dB signal shown below it is about 1% instead of 0.64% achievable otherwise (actually the second curve follows the Fletcher-Munson equal loudness contour).
Pedja
An externally hosted image should be here but it was not working when we last tested it.
Hello Bernhard,
According to Philips, with TDA1541A and with -60dBFS signal, THD (which means all the harmonics in sum) should be about -42dB and this distortion is referred to the level of the signal (i.e. to the -60dbFS). Under the same conditions S1 chips should have THD of -47dB. Both S1 and non S1 with 0dBFS signal are claimed to have -95dB THD. The S2 version is the same as S1 at low level but is claimed to have -97dB THD with 0dBFS signal.
Pedja
The measurement of TDA1541A you can see in my site is done using 0dBFS signal and sorry, I had a few times intention to put the notice about this and I did not do it till now but I promise I will do it tomorrow. So, the levels in the graph are not scaled with regard to the full level signal written in the CD. It is also important to be aware of this because of the I/V stage I use in this project. It is important to know that the distortion produced by this very I/V stage dominates here. It is also important to know that this distortion of this I/V decreases as the signal decreases, with -12dBFS it is about 0.04%). The distortion related to TDA1541A itself is here marginal. The real report about the DAC chip could be made using the opamp I/V and though I did try here I/Vs using a few opamps unfortunately I have not measured any of them. However, you can see below the distortion (0dBFS) of the same DAC with another common base I/V stage. Of course the distortion generated by the I/V still dominates but you can have idea about the sources of distortion involved here. (Btw the 50Hz with a few harmonics has nothing with the DAC but appeared as I connected to PC the cable of about 2.5m).Bernhard said:
An externally hosted image should be here but it was not working when we last tested it.
According to Philips, with TDA1541A and with -60dBFS signal, THD (which means all the harmonics in sum) should be about -42dB and this distortion is referred to the level of the signal (i.e. to the -60dbFS). Under the same conditions S1 chips should have THD of -47dB. Both S1 and non S1 with 0dBFS signal are claimed to have -95dB THD. The S2 version is the same as S1 at low level but is claimed to have -97dB THD with 0dBFS signal.
So, back on your measurements, I have read that and some results have sense but generally the results are too good (some are looking unbelievably low, say the two last ones) and looking at these measurements, now the measurement with 0dBFS signal you posted in this thread looks stranger. Do you have the idea why does K3 come at about -70dB? Does that player have the opamp I/V stage?Bernhard said:
Pedja
Pedja said:
Hi Pedja,
OT: Why don't you scale your analyzer to 0dB @ 1kHz ?
That would be much easier to read the graph and see @ what level the harmonics are.
I know that the distortion of the I/V dominates when a 0dB signal is used.
Also even if the I/V would be totally clean, with 0dB signal the distortion of the TDA would be invisible on an analyzer unless you have a 1 million $ machine.
The lowest signal my analyzer can detect is -75dB, with averaging. ***
So this is why I asked for the -60dB signal.
Because now the distortion of the TDA increases, dominates over I/V & preamplification, and is detectable with my analyzer.
The reason why my first measurements looked so bad, is that it was a -10dB sine from a test CD, and it was obviously a bad signal with high embedded distortion, maybe even recorded from an analog oszillator. But I think I wrote that in the posting.
On one test CD, you can even hear 50Hz noise in the pink noise :-(
I will be very happy to get the dithered 0dB signal, not for TDA testing, but for I/V stage testing.
So, why are some results from the -60dB signal so good ?
It is obvious that some plain 1541 chips beat some S1 chips.
Every chip got a different spectrum.
Please note: My analyzer is absolutely reliable, I always did 3 times measurement and the results was always within +/- 0,5dB for every harmonic.
+/- 0,5dB because of strong 50Hz noise whichs impact went up to 2kHz, and then shifted the boise floor up to -65dB :-(
I will get a Technics preamp, should fix that problem.
Unfortunately until now I could not see the K2 because it was below the -65dB noise floor.
Summary from my -60dB measurements:
All chips are different.
Some plain 1541 seem to be better than some S1.
1541A was worst of all.
SAA7220B was better in K2
SAA7220A was better in higher orders with some TDAs.
I am not sure if the SAA7220A works properly in my CD880, somewher I read one of the SAA need an additional signal on one pin.
This starts to be wicked.
IMHO for best sound it is necessary to select TDAs and try all combinations with SAA chips.
It seems that 1541A chips have always been selected, so if it is not S1. it is crap.
Plain 1541 has never been selected, so there might be some with S2 performance among them.
What is also important.
Philips maybe did selection only per total THD vs. frequency, if you look @ datasheet.
They did not look @ single harmonics.
And IMHO it is nothing worth if with SAA7220 B K2 is very very low, but K5 is higher.
Bernhard
*** The analyzer has only 12bit ADC but the repetitive accuracy is abolutely perfect.
And it got lots of features like storing results, adding, subtracting, dividing them from oter results, and lots of other options.
And I got it cheap on ebay.
Pedja said:
Hi Pedja,
OT: Why don't you scale your analyzer to 0dB @ 1kHz ?
That would be much easier to read the graph and see @ what level the harmonics are.
I know that the distortion of the I/V dominates when a 0dB signal is used.
Also even if the I/V would be totally clean, with 0dB signal the distortion of the TDA would be invisible on an analyzer unless you have a 1 million $ machine.
The lowest signal my analyzer can detect is -75dB, with averaging. ***
So this is why I asked for the -60dB signal.
Because now the distortion of the TDA increases, dominates over I/V & preamplification, and is detectable with my analyzer.
The reason why my posted graph looked so bad, could be:
I/V distortion dominates, like in the graph on your website.
Or the -0,5dB sine from test CDis a bad signal with high embedded distortion, maybe even recorded from an analog oszillator.
The -10dB sine from another test CD was terrible.
And on one test CD, you can even hear 50Hz noise in the pink noise.
I will be very happy to get the dithered 0dB signal, not for TDA testing, but for I/V stage testing.
So, why are some results from the -60dB signal so good ?
It is obvious that some plain 1541 chips beat some S1 chips.
Every chip got a different spectrum.
Please note: My analyzer is absolutely reliable, I always did 3 measurements of each chip and the results was always within +/- 0,5dB for every harmonic.
+/- 0,5dB because of strong 50Hz noise whichs impact went up to 2kHz, and then shifted the noise floor up to -65dB.
I will get a Technics preamp, should fix that problem.
Unfortunately until now I could not see the K2 because it was below the -65dB noise floor.
Summary from my -60dB measurements:
All chips are different.
Some plain 1541 seem to be better than some S1.
1541A was worst of all. Strong K5 !!!
SAA7220B was better in K2.
SAA7220A was better in higher orders with some TDA chips.
I am not sure if the SAA7220A works properly in my CD880, somewhere I read one of the SAAs needs an additional signal on one pin.
This starts to be wicked.
IMHO for best sound it is necessary to select TDAs and try all combinations with SAA chips.
It seems that 1541A chips have always been selected, so if it is not S1 then it is crap.
Plain 1541 has never been selected, so there might be some with S2 performance among them.
What is also important:
Philips maybe did selection only per total THD vs. frequency, if you look @ datasheet.
They did not look @ single harmonics.
And IMHO it is nothing worth if with SAA7220 B K2 is very very low, but K5 is higher.
Bernhard
*** The analyzer has only 12bit ADC but the repetitive accuracy is abolutely perfect.
And it got lots of features like storing results, adding, subtracting, dividing them from other results, and lots of other options.
And I got it cheap on ebay.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.