While I was comparing specs of the older AK4395 and the new flagship AK4399/90 DAC chips from AKM, I couldn't help but notice the big and strange difference in the digital filter specs. (just the DF specs, not overall thd+n and snr).
AK4395: 24bit, ±0.0002db, 110db
AK4399/90 : 32bit, ±0.005db , 100db
http://www.asahi-kasei.co.jp/akm/en/product/ak4395/ak4395_f01e.pdf
http://www.akm.com/datasheets/ak4390.pdf
http://www.asahi-kasei.co.jp/akm/en/product/ak4399/ak4399_f00e.pdf
So the old AK4395 has a 24bit filter and its ripple and attenuetion figures are way better than the newer AK4399/90 that has a 32bit filter.
Another interesting thing is that for the AK4395 the 110db figure stays the same for 96Khz and 192Khz measurements as well, whereas the 100db figure of the AK4399/90 deteriorates even further at 96Khz(95db) and 192Khz(90db)!
So what do you think is going on here ? It seems like the digital filter is inferior in the higher model ?
Ofcourse the bottom line is that the THD+N and SNR figures of the AK4399/90 are much better than the AK4395, but then again I am confused that if the stopband attenuetion is 90db then how can thd+n be -105db ?
AK4395: 24bit, ±0.0002db, 110db
AK4399/90 : 32bit, ±0.005db , 100db
http://www.asahi-kasei.co.jp/akm/en/product/ak4395/ak4395_f01e.pdf
http://www.akm.com/datasheets/ak4390.pdf
http://www.asahi-kasei.co.jp/akm/en/product/ak4399/ak4399_f00e.pdf
So the old AK4395 has a 24bit filter and its ripple and attenuetion figures are way better than the newer AK4399/90 that has a 32bit filter.
Another interesting thing is that for the AK4395 the 110db figure stays the same for 96Khz and 192Khz measurements as well, whereas the 100db figure of the AK4399/90 deteriorates even further at 96Khz(95db) and 192Khz(90db)!
So what do you think is going on here ? It seems like the digital filter is inferior in the higher model ?
Ofcourse the bottom line is that the THD+N and SNR figures of the AK4399/90 are much better than the AK4395, but then again I am confused that if the stopband attenuetion is 90db then how can thd+n be -105db ?
AKM's best DAC is the AKM4396 - Hands down. I have AKM eval boards for both the 4396 and the 4397. I've read their theories about their 32-bit products and met with one of their top engineers from Japan for an hour and heard the pitch. For me the results were not there.
A person cannot hear anything above 24 bits, and most normal people cannot hear above 20 bits.
The short of it is, stick with the 24-bit products. The 4397 was a step backwards in sound quality as compared to their 4396.
-David
A person cannot hear anything above 24 bits, and most normal people cannot hear above 20 bits.
The short of it is, stick with the 24-bit products. The 4397 was a step backwards in sound quality as compared to their 4396.
-David
I'm going to be posting reviews of all my eval boards on my website soon. I have a couple of board pics already posted. One of the boards is the AKM4396. It's a very nice board.
BTW, I'm selling my AKM4397 32-bit board when I write the reviews.
http://www.hifimaker.com/DAC-Eval.html
-David
BTW, I'm selling my AKM4397 32-bit board when I write the reviews.
http://www.hifimaker.com/DAC-Eval.html
-David
ok let me put it this way -
Forget about the 32bits. Lets just talk about the 24bit DACs. There are other 24bit DACs from AKM that have worse 'digital filter' specs than the AK4395. So the subject of discussion that I wanted to bring up was what is special about the digital filter in this DAC ? What is different ?
Forget about the 32bits. Lets just talk about the 24bit DACs. There are other 24bit DACs from AKM that have worse 'digital filter' specs than the AK4395. So the subject of discussion that I wanted to bring up was what is special about the digital filter in this DAC ? What is different ?
Hi Percy,
Sadly, the operation of the digital filter is more complex than this. The precision (number of bits) used for the arithmetic has no actual bearing on stopband attenuation - in an FIR filter, it's effects are on how much quantisation noise is generated, and the accuracy of the frequency response against the original design.
Indeed, more stopband attenuation isn't even always better. For a given length of filter, more stopband attenuation will require a gentler rolloff, which might not be desirable. Maybe in the 4399 they decided the attenuation didn't have to be more than 100dB, used a shorter filter (with the same rolloff as that in the 4395), and dedicated the saved silicon area to the modulator in order to get the extra THD performance. Without a lot more information at the system level, which they don't give in the datasheet, it's impossible to draw any conclusions from the numbers you've stated.
It's a shame, really. I've not seen any DAC datasheets that include enough measurements to objectively judge performance, but most do at least include the responses of their oversampling filters. Wolfson's range even go so far as to make a selling point of their responses, and TI do at least show the behaviour of the various modes. AKM just put a few numbers on theirs, and that's not enough to do anything with.
Sadly, the operation of the digital filter is more complex than this. The precision (number of bits) used for the arithmetic has no actual bearing on stopband attenuation - in an FIR filter, it's effects are on how much quantisation noise is generated, and the accuracy of the frequency response against the original design.
Indeed, more stopband attenuation isn't even always better. For a given length of filter, more stopband attenuation will require a gentler rolloff, which might not be desirable. Maybe in the 4399 they decided the attenuation didn't have to be more than 100dB, used a shorter filter (with the same rolloff as that in the 4395), and dedicated the saved silicon area to the modulator in order to get the extra THD performance. Without a lot more information at the system level, which they don't give in the datasheet, it's impossible to draw any conclusions from the numbers you've stated.
It's a shame, really. I've not seen any DAC datasheets that include enough measurements to objectively judge performance, but most do at least include the responses of their oversampling filters. Wolfson's range even go so far as to make a selling point of their responses, and TI do at least show the behaviour of the various modes. AKM just put a few numbers on theirs, and that's not enough to do anything with.
the only 'measurement' I have found so far is this - http://www.diyaudio.com/forums/showthread.php?s=&threadid=137703 (see first/second post).
there is also another thread by the same author on listening impressions of the two dacs.
there is also another thread by the same author on listening impressions of the two dacs.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.