It's an interesting question. The slowest sampling rate at 44.1KHz is about 23uSec. Values of transient recovery in I/V converters seems typically in the range of 50 to 100nSec. If we consider the fastest recovery at 50nSec at the slowest sampling rate of 23uSec (worst case condition of highest ratio) the ratio is 2 parts/ 1000 or about 500:1 (for 88.2KHz it becomes about 250:1)
If we consider this in terms as before, whereupon the "500" in this ratio is "a priori" good and the "1" is "a priori" bad, the good is 8 bits better than the "bad". This is to suggest that in order to achieve 16 bits of "good", the degree of the "bad" must be drastically reduced to support 16 bit resolution. Careful attention of transient conditions seems essential in I/V devices even under the most favourable of conditions.
To what extent do actual dac chips approximate ideal step current sources? Has it been measured, or is it something hypothesized?
We did the opposite, sort of, took a streamed hi-res digital source and recorded it with a nicely restored Otari R2R. Everyone present preferred the recording after it went through the tape recorder. In any case I'm a little confused, I thought DDD meant there is no master tape .
No dac chip can reproduce an ideal step current source, as being theoretically infinite in speed to the value of the current. Hence it becomes a relative comparative study between variant real world devices.
The TDA1541a settles from a full scale output to the least significant bit in 500nSec, or 1 part in 65,000, or to .000015 in 500nSec. If this follows an RC time constant path, the number of time constants at .63 reduction, or to 0.37 per time constant, this suggests about (0.37)^13 to .000025. Hence each RC time constant is about 500uSec/13 or around 40nSec. For 300uSec settling time as in the TDA1387, the RC time constant becomes 300uSec/13 or 23nSec. Perhaps someone can verify this.
Nevertheless, in testing of I/V networks for use with the TDA1541a, a normal digital function generator was used in its place, with a series 2K Ohm resistor feeding the inverting terminal. A square wave voltage set to +/- 4 volts pk/pk therefore simulates a +/- 2mA pk/pk into the I/V, as representing the full scale current of the 1541a. The signal generator has a fixed rise/fall time of 25nSec, hence seemingly a good starting point to evaluate I/V's in lieu of using the TDA1541a.
Those figures look to me rather optimistic. Checking a couple of datasheets, the AD811 (a video opamp, at the 'fast' end of the spectrum) says 65nS but this is only to 13bits (0.01%). However settling time decreases for smaller step sizes and those used in DSs are normally large (like 10V). AD797 (mid-range in the bandwidth spectrum but often seen in I/V duty in appnotes) looks to do around 400nS (to 16bits) for smaller step sizes (the graph doesn't go below 2V step).
Yes so there is no analog master tape, I mean in 2020 why would you need to record the bits onto tape?
...which is not the same as earlier where you said 'there is no master tape'. We all agree there's no analog master tape where the first letter is 'D'.
The SPARS code is a product of its time. Once upon a time digital tapes of all shapes and sizes roamed the earth.
I thought we agreed that there is no tape period these days. Digital tapes are full of many of the same problems as analog tapes.
Last edited:
One of my clients regularly receives classical music 'stems' from a label to
pass through his 16T Otari. Nothing else, just 1 pass through the R2R.
Added goodness 🙂
It was quite a PITA, due to the DR requirements of classical music I spent hours going through and solving many noise issues, lapped the heads etc etc.
He has since found a noise reduction unit and is happy. I'm not sure how the NR affects sound and don't want to ask. Tape machines are not my favorite
pass time but there just aren't many people left who can service them.
Having said all the above, the first time I heard a really good recording coming straight off a good 2" 16T machine (Ampex not Otari) I was quite
blown away by the sound. 3D, dynamic, clear and somehow riveting to listen to. I've rarely heard that with any type of digital.
I left the studio that evening with more questions than answers.
Great tape recorders can sound amazing. It's definitely part additive but it's
good.
TCD
I'm not sure if we are discussing the same thing. What I mean by "transient recovery" is not related to rise-times in relation to settling to any bit depth, rather to the time from a transient input overload, to the time when the signal on the inverting terminal returns back to a virtual ground, hence back to normal loop control. This can be considerably faster than the added time to settle.
It is believed that both the JVC4558 and NE5534 can recover from transients in under 100nSec, not on their own, rather that this necessitates a feedback capacitor from the output back to the inverting terminal. This seems likely dependant upon the magnitude of the output voltage as you suggest, being a function of feedback resistance and the full scale current output from the digital-to-analog converter.
It is also considered that the work you are doing as specific to filtering between the digital-to-analog device and the I/V can prevent such transient overload, whereupon the rate-of-rise of the input current signal has been slowed below the ability of the I/V to respond.
Allow me to add this almost endless source to the list
Internet Archive: Digital Library of Free & Borrowable Books, Movies, Music & Wayback Machine
George
Analog tape is still used sometimes. Its expensive to do compared to digital these days, so not everyone can afford to have access. The reason for using it at in the music business is exactly as you described, the sound is preferred in some cases. (The reason for using it as a possible ADC/DAC chain test source is simply for repeatability.)
Running a ITB mix though a 1/2" tape machine is a trick sometimes used in mastering. The better mastering houses can do almost anything it takes to polish turds into viable commercial releases (if the music is good enough). Sometimes tape is a thing that it 'takes.'
At one time there was a lot of interest in using very fast (say, video) opamps for I/V, perhaps for the reasons you state. However, fast opamp THD tends to be higher than that of commonly used audio opamps. Maybe 'fast' is okay for 16-bit audio.
For the kinds of dacs being produced today by ESS and AKM's AK4499, particularly the latter, they work very well with OPA1612. It may be that output slew rates from the dac are limited intentionally and or intrinsically. In any case, using OPA1612 it is possible to result in quite impressive measurements such as -124dB distortion.
If there is a problem with I/V accuracy due to settling then it doesn't appear to show up in any of the conventional measurements we usually rely on for evaluating dacs.
Last edited:
I have tried using devices like the LM7171 having a slew rate of 4000V/uSec, this with little success.
As you are likely aware, AKM has only recently begun making products with current output, hence I/V's are not normally used. For streaming I am using Mac/Tidal/ Audirvana fed via USB into the Earstudio ES100 containing a pair of AKM4375's. This is intimately connected to a designed amplifier/buffer stage before cabling to a tube preamplifier. For better or worse, this is one of the reference devices used to compare work on the I/V stage in the Naim CD3 player.
The OPA1612 looks an excellent device from the data sheets, though the topology at first glance looked problematic for I/V applications, appearing a current output
device with broadband high Ro. This view was countered by Fig. 28 on page 11 of the data sheets showing "open loop" Ro steadily dropping to below 1 Ohm, to about 10 Ohm at 2MHz, making this instead an excellent looking choice. The device also looks highly symmetric in form, as seemingly in compensation as well, as further to advance its use for I/V. To take advantage of these features, as specific to transient conditions in an I/V implementation, requires a capacitance connection from the output to the input. Fig. 35 on page 18, showing an I/V implementation, contains a 2200pF capacitor in the feedback path, considered vital to achieve the performance levels identified.
Few high speed CFA's can handle capacitors in the feedback path, with resistors alone too high to transfer and absorb input current transient energy in the output networks. CFA's are known to have low Ri on the inverting terminal, as typically 50 Ohm's in an AD844. This suggests of a proper normal function as well as transient energy absorption and recovery. The difference between transient absorption in the output vs. input, is that input transient energy directed to the output via capacitors feed into the power supplies, input transient energy absorbed by the input is fed into the active elements, the current mirrors, as thereupon subject to all manner of issues. This is to suggest that CFA's might appear like a good choice, rather they appear questionable in applications as I/V's.