Guys, this thread is about the RTX6001. Could you please start a new thread for the E-MU mod?
There is room for much better
Attached is an analog loop of AK4499 D/A and AK5572 A/D I'm working on. Absolutely no frills setup, no special low noise power supplies or reference voltages, clock is a 24.576 MHz quartz, Fin=1.5KHz, -1dBFS level, 22KHz bandwidth. Master clock 256*Fs, bit clock=64*Fs and L/R clock are all generated by an AK4118 DIR, I/O is SPDIF/optical. Op amps are AD1612 for the D/A and LM49710 for the A/D.
Third harmonic is -127dB (the worse channel, the better channel is slightly under -130dB), everything else is close to the (under) -150dB noise floor. Extra 2-3dB improvement is expected for Mono mode. I'm not convinced the third harmonic is not a setup artifact, though, I was expecting the 2nd harmonic to be dominant., will have to look closer. No idea what low noise power supply and reference voltages plus an ultra low phase noise clock can bring, I suppose not much from this point.
Attached is an analog loop of AK4499 D/A and AK5572 A/D I'm working on. Absolutely no frills setup, no special low noise power supplies or reference voltages, clock is a 24.576 MHz quartz, Fin=1.5KHz, -1dBFS level, 22KHz bandwidth. Master clock 256*Fs, bit clock=64*Fs and L/R clock are all generated by an AK4118 DIR, I/O is SPDIF/optical. Op amps are AD1612 for the D/A and LM49710 for the A/D.
Third harmonic is -127dB (the worse channel, the better channel is slightly under -130dB), everything else is close to the (under) -150dB noise floor. Extra 2-3dB improvement is expected for Mono mode. I'm not convinced the third harmonic is not a setup artifact, though, I was expecting the 2nd harmonic to be dominant., will have to look closer. No idea what low noise power supply and reference voltages plus an ultra low phase noise clock can bring, I suppose not much from this point.
Attachments
Last edited:
I would guess a marginally better looking plot, in meaningful terms probably nothing. Pretty impressive right as it stands I'd say.
What is the performance like at higher sampling rates (say 88.2kHz) where the performance generally seems to deteriorate a bit IMVLE?
With the RTX I generally stick to 48kHz 24 bit unless I need more bandwidth. This seems to be a sweet spot performance wise. I have not done any of the retrofits although I have the kits to do so. Pragmatically speaking the performance is good enough most of the time. I will put the shields on at some point to deal with the cross talk issue.
What is the performance like at higher sampling rates (say 88.2kHz) where the performance generally seems to deteriorate a bit IMVLE?
With the RTX I generally stick to 48kHz 24 bit unless I need more bandwidth. This seems to be a sweet spot performance wise. I have not done any of the retrofits although I have the kits to do so. Pragmatically speaking the performance is good enough most of the time. I will put the shields on at some point to deal with the cross talk issue.
Correct, at these levels everything is just a number. Didn't get yet to higher Fs but preliminary the remarkable thing is the lack of increasing noise floor and higher frequencies.
The limitation in this setup is (of course) the A/D. The AK4499 D/A is absolutely stunning, has around -130dB THD+N, better than the data sheet spec.
The lack of elevated noise floor at high frequencies probably indicates they're doing something clever with the noise shaping. I've seen some ugliness at high sample rates where the noise floor is 20 - 30dB worse above 22kHz than I would have expected. My last measurement set up based on a M-Audio 24192 seemed to have this problem whereas its predecessor a 2496 definitely didn't.
Sadly, the RTX is dead and gone, if this is something that might benefit the readership I'd be delighted if you started a new thread and shared details.
Sadly, the RTX is dead and gone, if this is something that might benefit the readership I'd be delighted if you started a new thread and shared details.
Unfortunately, a new "RTX" by far exceeds my capacity and resources (in particular time). So I have to constrain myself to this R&D evaluation phase...
To add insult to injury, the sad RTX story clearly shows there is no money in such large a project, so IMO it will remain for the foreseable future a rare gem of engineering.
After I'll get more results I'll perhaps open a new thread. After completely characterizing the A/D and D/A the next steps are to add an XMOS based interface, then a BF561 Blackfin DSP I have here, waiting for an interesting audio related project.
Last edited:
The limitation is most likely primarily the ADC (AK5572).
I assume that you go digital output -> AK4499 -> AK5572 -> digital. Is that correct?
It would be interesting to see a measurement with a notch filter, to get the ADC performance out of the equation.
A number of AKM converters show a flat noise floor up to the highest frequencies. That was one of the reasons for choosing them. Apart from the very low distortion, primarily from the AK5394A of course.
The AK5572 does have an elevated noise floor above 20 kHz though.
It is correct that the RTX6001 has been discontinued. But perhaps it will get a second life ?
I am working on an improved version (back to being a hobby project).
I have also considered the AK4499, although it is very expensive. Perhaps I will settle for a slightly lower performance, but still improved compared to the RTX6001, primarily on the generator side.
Yes, what I said above.
Yes.
Time permitting... I have a notch filter 86dB deep around 1KHz, but it is single ended... AK4499 D/A to AK5572 A/D analog connection is balanced...
Indeed AK4499 is very expensive and, I would say, it's a "because I can" piece of audio engineering. It may have a market outside of the realm of audio, though, like instrumentation.
Last edited:
Cross talk
Perhaps you have noticed my issues about cross-talk of harmonics between the channels of the ADC (see: DiAna, a software Distortion Analyzer). So you might consider using two separate chips for the left and right ADC channel in the next version of the RTX (this also could lower the noise floor by 3dB if you sum pairwise the ADC outputs together, I guess).
Cheers, E.
Hi Jens,
Perhaps you have noticed my issues about cross-talk of harmonics between the channels of the ADC (see: DiAna, a software Distortion Analyzer). So you might consider using two separate chips for the left and right ADC channel in the next version of the RTX (this also could lower the noise floor by 3dB if you sum pairwise the ADC outputs together, I guess).
Cheers, E.
I have started something similar, really as the baseband implementation of a
phase noise test set, but for that I have found a fully digital solution that works
at the expense of $$$.
It is centered around a BeagleBoneBlack, a 1 GHz ARM RISC with floating point
and running Debian Linux. All software editing & compilation can be done
on the BBB, you simply ssh into it from your laptop. I could compile fftw,
the fastest FFT in the west on the ARM.
There is a mutation of my control program for the 89441A FFT analyzer
that runs on a PC and can do GPIB / IEEE488-like conversation over the LAN.
The LAN interface of the BBB is isolated via transformers.
No ground loops. WLAN should be possible also, but I did not explore that.
The ADC is a LTC2500-32. It delivers up to 1 MSPS @ 24+ bits. Sample rate is
programmable in steps of 2 and derived currently from a 100 MHz oscillator.
It uses 100 MHz SPI for data transfer. Smaller sample rate delivers more bits.
Digital filters and decimation are inside the LTC2500-32.
In addition to the BBB, one currently needs 2 1-square-inch boards, home-etched
in my case. One is a Coolrunner2 CPLD that creates the sample rate and
converts the 100 MHz SPI to a slower byte-wide interface.
The other is the ADC board. It has the ADC, LT3042 regulators for analog and
digital VCC, the recommended differential driver for the ADC and another
regulator for the negative driver supply. Reference voltage is provided by
a LT6655.
The BBB can handle upto 3 ADCs at full sample rate. That was important for me
so I can calculate the 3-cornered hat. (A cross correlation that can separate
signals/noise from 3 sources. Allows to dig deeper into the noise.)
The BBB has 2 on-chip IO processors with 5 nsec predictable instruction rate.
They are programmable in C , the necessary compiler is included on the BBB.
One of these IO processors fetches the ADC data bytewise from the CPLD
and deposits them as samples in a 16 Kword shared on-chip static RAM, where
ARM can fetch them away. I have implemented a ping-pong buffer. Time series
can be megabyte-large.
The shared ram serves also for command handshaking. The control program
on the laptop can set for example the decimation factor, pass it as GPIB-like
string over the LAN to the ARM in the BBB. The ARM deposits it in a command
register in the shared RAM and the PRU IO-Processor then sets up the ADC.
That works already. I cannot yet transfer ADC buffers larger than the shared
RAM, that still needs to be programmed. Project overload.
I would create a repository on github or sth. like that with the Altium files,
PC, ARM + PRU programs and the boot image if there is real interest.
That thing would be a nice self-contained FFT analyzer without USB drivers
and such. Just open port 5010 on 192.168.178.38 and write ascii commands
or read time samples or spectra. A BBB costs $50 or so.
regards, Gerhard
phase noise test set, but for that I have found a fully digital solution that works
at the expense of $$$.
It is centered around a BeagleBoneBlack, a 1 GHz ARM RISC with floating point
and running Debian Linux. All software editing & compilation can be done
on the BBB, you simply ssh into it from your laptop. I could compile fftw,
the fastest FFT in the west on the ARM.
There is a mutation of my control program for the 89441A FFT analyzer
that runs on a PC and can do GPIB / IEEE488-like conversation over the LAN.
The LAN interface of the BBB is isolated via transformers.
No ground loops. WLAN should be possible also, but I did not explore that.
The ADC is a LTC2500-32. It delivers up to 1 MSPS @ 24+ bits. Sample rate is
programmable in steps of 2 and derived currently from a 100 MHz oscillator.
It uses 100 MHz SPI for data transfer. Smaller sample rate delivers more bits.
Digital filters and decimation are inside the LTC2500-32.
In addition to the BBB, one currently needs 2 1-square-inch boards, home-etched
in my case. One is a Coolrunner2 CPLD that creates the sample rate and
converts the 100 MHz SPI to a slower byte-wide interface.
The other is the ADC board. It has the ADC, LT3042 regulators for analog and
digital VCC, the recommended differential driver for the ADC and another
regulator for the negative driver supply. Reference voltage is provided by
a LT6655.
The BBB can handle upto 3 ADCs at full sample rate. That was important for me
so I can calculate the 3-cornered hat. (A cross correlation that can separate
signals/noise from 3 sources. Allows to dig deeper into the noise.)
The BBB has 2 on-chip IO processors with 5 nsec predictable instruction rate.
They are programmable in C , the necessary compiler is included on the BBB.
One of these IO processors fetches the ADC data bytewise from the CPLD
and deposits them as samples in a 16 Kword shared on-chip static RAM, where
ARM can fetch them away. I have implemented a ping-pong buffer. Time series
can be megabyte-large.
The shared ram serves also for command handshaking. The control program
on the laptop can set for example the decimation factor, pass it as GPIB-like
string over the LAN to the ARM in the BBB. The ARM deposits it in a command
register in the shared RAM and the PRU IO-Processor then sets up the ADC.
That works already. I cannot yet transfer ADC buffers larger than the shared
RAM, that still needs to be programmed. Project overload.
I would create a repository on github or sth. like that with the Altium files,
PC, ARM + PRU programs and the boot image if there is real interest.
That thing would be a nice self-contained FFT analyzer without USB drivers
and such. Just open port 5010 on 192.168.178.38 and write ascii commands
or read time samples or spectra. A BBB costs $50 or so.
regards, Gerhard
Attachments
Last edited:
Keep in mind that the freq. response needs to be adjusted (flatten by DSP) as seen in the RME ADI-2 Pro fs.
They are using one AK5574 ADC in Stereo mode. May consult the manual & graphs
Hp
Keep in mind that the freq. response needs to be adjusted (flatten by DSP) as seen in the RME ADI-2 Pro fs.
They are using one AK5574 ADC in Stereo mode. May consult the manual & graphs
Hp
It probably doesn't "need" to be corrected if you are using it for audio. For measurements, I agree.
You still working on the power supply change for the rtx6001? Also will the newer generators be drop in for the rtx6001?
Yeah, it's a little worse. The UPD is a rather old piece of equipment, but at it's time it was well over anything that AP was offering (System One).