Haha - welcome to the world of oscillator subjective performance Andrea,
nicely done.
It will be interesting to compare measurements of these oscillators and see
what actually separates them objectively.
Did Crystek have same isolated PS as others?
Terry
Very good observer. That's very important. The on-board 3.3V regulator looks like a 1117-3.3 to me.
I have tested the Cristek in both situations: on-board Aune regulator and separate TWRPS-pp micro regulator, the same used for the Pierce oscillator. There are no substantial difference, the Cristek sounds harsh and confused. Only the TWTMC Driscoll All in One uses a different shunt regulator on-board.
As I said, I think the crystal makes the difference in every case. The Cristek uses a very tiny crystal, that's not a good way. I believe that's the reason why the Cristek sound so bad and the Driscoll with a budget "large" HC-49/U crystal sounds very nice, not far from the top performers.
BTW, the phase noise plots of the Cristek and the Driscoll predicts, at least in part, the sonic results.
For the rest of the measurements you have to wait, now I cannot access the Agilent gear. The Driscoll was measured by TFC in Scotland, but I cannot ask them to measure all the oscillators.
Maybe, if someone would design a decent PCB for the Wenzel design, I subscribe immediately...
Since I have many requests, the All in One PCB will be available soon (obviously with schematic).
Well, If someone does the cross correlation software in a somewhat hardware independent
way, I'll do the board. With "somewhat hardware independent" I mean that it runs on a
stereo sound card AND a Fourier analyzer. In fact, I have done it already. Andrea found
it to complicated. But it was Wenzel 1:1, just the switches replaced by relays so it could
be computer controlled, and the input amp has some more FETs to further reduce the noise
and avoids an unobtainium FET.
A pic of the existing board is somewhere earlier in this thread.
My new preamp based on IF3602 is simpler.
Also, the ring mixer and low pass filter was already on the board.
But I won't invest any time in a thing that does no x correlation, that would be
only lost time for me. I already have access to an E5052B. I can live with a clumsy
thing that needs two reference oscillators, but never with less dynamic range.
regards, Gerhard
Last edited:
Hi Gerhard,
do you mean software for PC or firmware? If you mean software for PC, I could develop it but not hardware or platform independent. My knowledge is limited to C#, Vb.net and Delphi, so I can develop for Windows only.
Can you explain better about the software and the soundcard? What soundcard are you thinking to use?
If my skill was enough I could start a new thread for a project group.
Andrea
HPW software has the cross correlation already. He added it back when I was planning to make a simpler version of the hardware. My intern found a job so that project stalled. Gerhard's board is interesting but not a trivial undertaking to build. Maybe a combo project would resolve this going forward. With cross correlation the reference oscillators do not need to be heroic but they do need to be pretty good. I was hoping that the Crystek's would be good enough. Maybe the Driscoll board could handle that task. Especially if we can add some VCO capability.
Hi Demian,
no problem to add VCO capability to the Driscoll oscillator. Maybe it could be part of the new PCB (if Gerhard subscribes the project...)
Realistically, for audio we only need a limited number of frequencies so that simplifies the process. Some of the switching and other stuff in the Wenzel is for aspects we don't need in this.
We need two oscillators, two good mixers, two good low noise amplifiers. The amplifiers are the biggest question for me. What can we do to simplify the amp requirements? How low does the noise need to be? Across what frequency range? There needs to be DC amplification for the PLL as well. Does that need to be chopper stabilized? They need to match in gain pretty well and be pretty low noise. For mixers I was thinking of using some from Minicircuits that seem well suited to this: http://www.minicircuits.com/pdfs/SYPD-1+.pdf
The power requirements do not need to be as wide as the Wenzel needed.
I have not tried HPW's software Home lately. I did try the demo version but it works so differently from my familiar world I had to pass on it then.
We need two oscillators, two good mixers, two good low noise amplifiers. The amplifiers are the biggest question for me. What can we do to simplify the amp requirements? How low does the noise need to be? Across what frequency range? There needs to be DC amplification for the PLL as well. Does that need to be chopper stabilized? They need to match in gain pretty well and be pretty low noise. For mixers I was thinking of using some from Minicircuits that seem well suited to this: http://www.minicircuits.com/pdfs/SYPD-1+.pdf
The power requirements do not need to be as wide as the Wenzel needed.
I have not tried HPW's software Home lately. I did try the demo version but it works so differently from my familiar world I had to pass on it then.
I have personally not the least interest in using a sound card, but it would be needed
to get a somewhat broader user base. So, the interface "under the hood" could be
a standardized file format. I have bought an Agilent 89441A that solves already some
of these problems, it can do crosscorrelated FFTs, for example. But still it is an
awful lot of work to do. It took me more than a month of my leisure time to control
it over the network to do 7 FFTs over 7 decades, combining everything together
and feeding it into gnuplot to get a nice combined picture from 0.1 Hz to 1 MHz.
Communication over files should be ok, this kind of measurement usually
takes minutes to hours because of the extensive averaging that is required.
It is far away from "real time".
The final problem is that you get numbers, but have no way to verify them.
I just have made a box that produces noise from a few KHz to 150 MHz
that should be reproducible to within a dB (thermal resistor noise + LMH6702s)
and that noise can be added to an oscillator signal with power combiners.
(Mini Circuits PSC2-1 or such.) That should provide a known reference line at
say, -150 dBc.
Listening session again
I have just done one more test with another oscillator that confirm the previous session.
- TWTMC-P Pierce with budget HC-49/U AT-Cut crystal from TFC powered by the on-board Aune regulator
2) Score: 6+. Comparing the result with the TWTMC-P cold welded HC-43/U AT-Cut crystal from Laptech and isolated power supply, the difference is clearly audible. Bass are less dumped, vocals have lower thickness, upper octaves are less smooth. The soundstage is not so wide and deep as the big Pierce, unplugged music is less realistic. BTW it remains detailed and never harsh and confused like the Cristek. You can listen without getting tired after a few minutes how happens with the Cristek. Obviously, the differences are not shades, but it never becames harsh and aggressive and keeps a decent level of detail, while the Cristek sounds always confused, with lack of detail and harsh character. All in all the result is appreciable, since it uses a cheap crystal and a questionable power supply.
Again the crystal makes the difference. The budget HC-49/U TFC crystal has anyway a Q around 90k and ESR less than 5 ohm, that's a remarkable result for a cheap resistance welded crystal. I wonder what is the Q of tiny crystal used in the Cristek oscillator, I suspect very low, and I suspect that explains the sonic results.
I have just done one more test with another oscillator that confirm the previous session.
- TWTMC-P Pierce with budget HC-49/U AT-Cut crystal from TFC powered by the on-board Aune regulator
2) Score: 6+. Comparing the result with the TWTMC-P cold welded HC-43/U AT-Cut crystal from Laptech and isolated power supply, the difference is clearly audible. Bass are less dumped, vocals have lower thickness, upper octaves are less smooth. The soundstage is not so wide and deep as the big Pierce, unplugged music is less realistic. BTW it remains detailed and never harsh and confused like the Cristek. You can listen without getting tired after a few minutes how happens with the Cristek. Obviously, the differences are not shades, but it never becames harsh and aggressive and keeps a decent level of detail, while the Cristek sounds always confused, with lack of detail and harsh character. All in all the result is appreciable, since it uses a cheap crystal and a questionable power supply.
Again the crystal makes the difference. The budget HC-49/U TFC crystal has anyway a Q around 90k and ESR less than 5 ohm, that's a remarkable result for a cheap resistance welded crystal. I wonder what is the Q of tiny crystal used in the Cristek oscillator, I suspect very low, and I suspect that explains the sonic results.
Attachments
Hi all,
In the quest to obtain better sonics via better clocking.
Is the lowest short term phase noise most critical?
How about the absolute requency, how important is it ro obtain say 11.2897mhz.
If a particular crystal with excellent phase oise is off say 11.200 mhz or 11.300mhz.
How does this affect the sonics.
If the frequecies can be pulled to exact accuracy via voltage to pin 1 of a ocxo or vcxo,how will we acheive this when power supply voltages are changing overvthe course of the day assuming we are using adjustable regulators and a trim pot to get this exact frequency.
In the quest to obtain better sonics via better clocking.
Is the lowest short term phase noise most critical?
How about the absolute requency, how important is it ro obtain say 11.2897mhz.
If a particular crystal with excellent phase oise is off say 11.200 mhz or 11.300mhz.
How does this affect the sonics.
If the frequecies can be pulled to exact accuracy via voltage to pin 1 of a ocxo or vcxo,how will we acheive this when power supply voltages are changing overvthe course of the day assuming we are using adjustable regulators and a trim pot to get this exact frequency.
Attachments
Last edited:
Gerhard:
The Agilent analyzer is a pretty powerful box. No wonder it took you quite a while to figure out controlling its feature set.
For audio I'm not convinced that noise above 100 KHz is all that important. it is for other applications. The importance of close in noise is what is being explored here so response to 10 Hz or less is important. In this context a soundcard would be adequate.
I would think that adding the extra components to Andrea's dual Driscoll to implement the block diagram should not be too difficult, especially if the noise floor of around 1 nV/rtHz of an LME49990/AD797/LT1024 would be low enough.
remaining questions on this-
1) reference noise source- couldn't a voltage modulated oscillator provide a suitable reference if all the details are worked out? A sinusoidal modulation with peak voltage matching the frequency offset from a known voltage could be translated into a known phase noise level? Getting some reference is important. I think there is something in the Wenzel lit on that issue as well. They also have a procedure for manual calibration that is not real easy to follow on paper.
2) Could we start with 24.568/22.5792 MHz and a multiplier/divider for all the variations? or do we need discrete oscillators for each frequency?
3) Do we need an input signal conditioning stage? A limiting amplifier to get the necessary level and remove amplitude noise?
Are you setup to plot the Jtest output of a DAC with the different oscillators? I think it would be really interesting to see if it shows anything meaningful.
- Status
- Not open for further replies.