I found recently even "mother" of dac TDA1540 14bit version, sounds very good really.
Working in NOS mode up to 192KHz sampling rates.
(Without SAA digital filter... Amanero I2S input + CPLD packing needed input format in NOS mode + Riv + JFET gain stage and output buffer)
Very simple set-up just to sniff the sound...
Working in NOS mode up to 192KHz sampling rates.
(Without SAA digital filter... Amanero I2S input + CPLD packing needed input format in NOS mode + Riv + JFET gain stage and output buffer)
Very simple set-up just to sniff the sound...
Hi.
I guess I wrote a little too fast...
Of course can the problem be taken care of in the digital domain - hence the existence of digital OS filters - you're right.
Skipping the digital filter necessitates very steep analogue filters that require time, effort and more importantly - competence to design, specially if
the designer aims to make them sing.
The usual shelved-HP filters, often seen on the NOS Dac outputs really don't solve anything.
As mentioned earlier, I think the problem of some individuals not finding the NOS designs sonically satisfying might be more neurological than technical.
It appears some people have auditory systems that are, for some unknown reason, sensitive to the NOS artifacts that are not
harmonically related to the music.
I did my share of NOS-making in the 90's.
Although the low fs sounded amazing, there was a very bad, unnatural quality to the rendition that always made me
question - my mods at first, and then the NOS idea later on.
The R-ladder Dacs are interesting in that they can do without the high clock rates necessary for bitstream modulation.
Cutting down on the clock frequency leads to lower jitter, lower noise, lower radiation, more effective grounding, less glitching, more effective PSUs, less .....
I have found all R2R dacs to be uncomfortable with higher clocks and bit-rates.
This might be one of the reasons behind the NOS sex-appeal => bringing everything down to NOS improves the internal circuit performance of multibit dacs.
With all that being said, I would like to name a DAC made in the 90's by a smart swedish engineer - using the CS4328 BITSTREAM dac.
Although it required a very good transport (for obvious reasons), its sound quality was nothing short of breathtaking.
Nothing I have heard in my life has been able to induce the same addictive sensation as when I listened to this one, not even the
TDA1541a inside cd-77 by T. Loesch.
Attachments
Last edited:
Actually a very good strategy, moving all the signal processing to a unit
far more capable than any embedded solutions or single chips...not to mention the SW flexibilty it provides.
I used the CS4329 dac chip in my car in the latest 90's, it sounded surprisingly fine, but the TDA1541A is another story.
The CS4329 is not the same as the CS4328. The CS4328 was very good but limited to 48 KHz max sample rate and no support for HDCD. I manufactured a number of DACs (the Entec Number Cruncher) using that combo. So did Forsell. John Curl's partner at the time use my boards in a Blowtorch level DAC that was also well respected. Our trick was to use the voltage out from the chip + a passive filter. The results were extraordinary.
The later chips were not by the same team. They all became executives at Crystal and moved on. Crystal was absorbed into Cirrus. The underlying technology lives on at Cirrus (no new stuff for 10 years plus in this category) and at AKM whose latest stuff (AK4490 and AK4497) are the best I have tried and worked with.
Ladder DAC's need exceptional sample and hold circuits to work. Both timing (aperture window) and voltage accuracy need to be really good to work. The timing needs to be at the same picosecond performance as the master clock or you did not get want you were after. For 16 bit performance the voltage and the timing need to be 96 dB accurate.
The Cirrus/AKM voltage out switched cap DACs are the least sensitive to jitter.
Brick wall antialising filters are doable in the analog domain but quite difficult. The usual VCVS active filter will be stressed with the common mode voltages at the input. Especially if the ultrasonic components are so close to the audio range.
The later chips were not by the same team. They all became executives at Crystal and moved on. Crystal was absorbed into Cirrus. The underlying technology lives on at Cirrus (no new stuff for 10 years plus in this category) and at AKM whose latest stuff (AK4490 and AK4497) are the best I have tried and worked with.
Ladder DAC's need exceptional sample and hold circuits to work. Both timing (aperture window) and voltage accuracy need to be really good to work. The timing needs to be at the same picosecond performance as the master clock or you did not get want you were after. For 16 bit performance the voltage and the timing need to be 96 dB accurate.
The Cirrus/AKM voltage out switched cap DACs are the least sensitive to jitter.
Brick wall antialising filters are doable in the analog domain but quite difficult. The usual VCVS active filter will be stressed with the common mode voltages at the input. Especially if the ultrasonic components are so close to the audio range.
With its many psu pins, many clock inputs, and not to mention 19 decoupling caps, TDA1541 provides endless tweaking possibilities.
As 1audio correctly pointed out CS4328 was limited to 48 kHz, and featured a fixed OP on the output that could not be by-passed.
Nevertheless, sonically it was a legendary chip - the best Crystal ever did imho (thanx 1audio for the unside info, interesting read).
I talked to Peter (the engineer behind the Bremen Dac) when I was in Stockholm many years ago.
He found the chip the best one on the msrket then.
He was also the engineer behind Forsell Air Reference dac.
The only problem in his dac was that it featured no provisions against jitter.
The whole da was clocked by the extracted MCK from Rx chip.
He mentioned that at the time technology did not offer transparent solutions to reclocking as the best ones sounded similar to no reclocking - at best.
He did something similar to Lampizator to fight jitter:
Took a cdm9 based cd player, massively cleaned up the psus, and upgraded the chips to the fastest ones at the time, and installed a 3000 USD (early 90s!!) Vectron 11.2896 clock as the reference generator.
Needless to say that the combo is still my reference to this date...
I found recently even "mother" of dac TDA1540 14bit version, sounds very good really.
Working in NOS mode up to 192KHz sampling rates.
(Without SAA digital filter... Amanero I2S input + CPLD packing needed input format in NOS mode + Riv + JFET gain stage and output buffer)
Very simple set-up just to sniff the sound...
Sometimes in audio to go ahead you need to look back. At the beginning preferably.
Think about loudspeakers. The most incredible and realistic sound I've ever heard was through a old WE 555 field coil compression driver.
On dacs and the TDA1540. Some of the very best sounding present commercial dacs only have a hair more than 14 bits resolution. That's theory, reality is likely a bit worse.
A really curious parallel.
Ops... and what about tubes, such as the 845, developed in the very early thirties?
It is great nonsence to use an oven-controled Xtal oscillator in digital audio.
Herbert.
First question: is the temperature sensitivity drift a source of close in phase noise that is the core question this is whole effort intended to address?
Second question: Is the oven good enough (stable enough) to suppress the variations from temperature? Good ovens are .1 degree or better but not homebrew.
The easy check would be listening. However it really should be blind.
Sent from my LG-H811 using Tapatalk
Second question: Is the oven good enough (stable enough) to suppress the variations from temperature? Good ovens are .1 degree or better but not homebrew.
The easy check would be listening. However it really should be blind.
Sent from my LG-H811 using Tapatalk
You can see it quite often, that for a given oscillator the optimum phase noise
is not at the intended frequency, and both wander all over the place with
varying temperature. Also, there are so called activity dips that move over
temperature and excitation. Having at least constant temperature reduces
the searching room by one dimension.
Good ovens are measured in MilliKelvin.
The FEI405 precision oscillator has one crystal on its favourite frequency,
whatever that may be exactly, for long-term stability and another crystal
osc on 15.000000 MHz (for the one specimen I happen to own) that is
slowly locked to the first one via a hi-res DDS.
That one goes down to E-13 land for time intervals from a second to a minute.
You seldom get EVERYTHING right in the same oscillator:
voltage dependency, close-in phase noise, noise floor, level,
long-time stability, tuning range, temperature range, activity dips...
regards, Gerhard
Last edited:
And especially cost.
That's an interesting way to improve performance but probably with some other surprises that will bite when you least expect it (like odd spurs from the DDS). Sounds a little like one of their Rubidiums with a crystal instead of a physics package.
I still am not convinced that the close in phase noise is a significant culprit for audio performance. And its possible that the improved performance Andrea noticed could be from simply a better drive for the clock distribution or possibly slower transitions causing less EMI. Its difficult to separate the pieces to determine what, if anything, makes a difference. If the crystal is the difference great, but if a circuit that has slower rise and fall times gives better results then its not hard or expensive and doesn't need premium crystals. I'm all for doing things right if its no harder or more expensive than doing it poorly. Usually its just more work.
That's an interesting way to improve performance but probably with some other surprises that will bite when you least expect it (like odd spurs from the DDS). Sounds a little like one of their Rubidiums with a crystal instead of a physics package.
I still am not convinced that the close in phase noise is a significant culprit for audio performance. And its possible that the improved performance Andrea noticed could be from simply a better drive for the clock distribution or possibly slower transitions causing less EMI. Its difficult to separate the pieces to determine what, if anything, makes a difference. If the crystal is the difference great, but if a circuit that has slower rise and fall times gives better results then its not hard or expensive and doesn't need premium crystals. I'm all for doing things right if its no harder or more expensive than doing it poorly. Usually its just more work.
So I am [not]. I'm interested in ultra-stable oscillators for space and other time keepig
applications, but not for audio. Just imagine the damage done by a LP being off-center
by 0.1 mm, maybe because the hole is bigger than the center peg. Oh, those FM and
PM artefacts, and nobody complained in the last 100 years!
cheers, Gerhard
Gerhard,
I thought we were treating clock oscillators for digital audio, right?
The LP... good point! I remember me that I was surprised by the first (bad) CD-players because a piano no longer whined!!!
But this has nothing to do with clock jitter. For whining the jitter should exeed milliseconds at subsone frequencies. Here we are talking about picoseconds.
For a digital audio clock oscillator the absolute frequency is of very little importance so does the long term stability.
Cheers,
Herbert.
- Status
- Not open for further replies.