I got most of that from the writeup I saw.
At this stage we would like to see which analog multiplier they used and how its configured. That is the biggest impediment to distortion in any of these designs. The MDAC uses a resistor array with fet switches to change gain which is pretty linear as Davids oscillator has demonstrated. Otherwise its FET's, analog multipliers or optocouplers all of which have significant nonlinearity. With the best common analog multipliers getting .25% accuracy the output would need 40+ dB attenuation to get to low distortion as we call it. I'm sure this is a misrepresentation but the concept is correct.
At this stage we would like to see which analog multiplier they used and how its configured. That is the biggest impediment to distortion in any of these designs. The MDAC uses a resistor array with fet switches to change gain which is pretty linear as Davids oscillator has demonstrated. Otherwise its FET's, analog multipliers or optocouplers all of which have significant nonlinearity. With the best common analog multipliers getting .25% accuracy the output would need 40+ dB attenuation to get to low distortion as we call it. I'm sure this is a misrepresentation but the concept is correct.
Check out the Prism audio dscope - a fabulous product IMO. Cheaper new than a lot of the second hand ap stuff and fabulous software. Support has been excellent too - they wrote custom scripts for me to calculate even/odd harmonics in thd for example. No affiliation other than being a customer. I also have ap gear and use it but most of my work is now done on the ds3.
Or maybe the new SR1+:
http://www.thinksrs.com/downloads/PDFs/ApplicationNotes/What's new with SR1.pdf
http://www.thinksrs.com/downloads/PDFs/ApplicationNotes/What's new with SR1.pdf
Trying to measure distortion of an MDAC multiplier is an interesting endeavor.
The clacking of the mos switches do a real number on a TT active filter. All sorts of noise lobes are created. At the valley of the lobe nodes is the true distortion. But the noise is not representative of what's actually output from the Mdac IV convertor.
The multiplier core for my generator attenuates the Mdac output by 10:1 and the output of the multiplier is decoupled from the oscillator by 20dB. The Mdac sees 40dB of attenuation to the oscillator core.
Using Mdacs for tuning is problematic. One Mdac has a maximum output capacitance at full code of 200pF. The output capacitance of the Mdac is code dependent. I have two in parallel per section. This makes up 400pF at full code at the input to the integrators. The integrator capacitor has to be greater than the capacitance seen at the integrator input or the oscillator level is unstable and the multiplier's authority is exceeded. Consequently, the range capacitor value is limited to a scale larger than what is optimum for lowest distortion. The greater the integrator cap value the greater the capacitance load is on the op amp and the distortion rises. On the other hand the larger the integrator cap the lower the noise is.
I,m seriously considering redesigning the oscillator with a relay switched resistor string.
This will solve the problem with the Mdac parasitic capacitance. It will increases the the loop stability and extends the frequency range of the oscillator and the distortion will be lower. In the end using relay is no more expensive that using Mdacs and might even be a bit less expensive.
The multiplier and PI controller will still use Mdacs because this is where they shine.
The clacking of the mos switches do a real number on a TT active filter. All sorts of noise lobes are created. At the valley of the lobe nodes is the true distortion. But the noise is not representative of what's actually output from the Mdac IV convertor.
The multiplier core for my generator attenuates the Mdac output by 10:1 and the output of the multiplier is decoupled from the oscillator by 20dB. The Mdac sees 40dB of attenuation to the oscillator core.
Using Mdacs for tuning is problematic. One Mdac has a maximum output capacitance at full code of 200pF. The output capacitance of the Mdac is code dependent. I have two in parallel per section. This makes up 400pF at full code at the input to the integrators. The integrator capacitor has to be greater than the capacitance seen at the integrator input or the oscillator level is unstable and the multiplier's authority is exceeded. Consequently, the range capacitor value is limited to a scale larger than what is optimum for lowest distortion. The greater the integrator cap value the greater the capacitance load is on the op amp and the distortion rises. On the other hand the larger the integrator cap the lower the noise is.
I,m seriously considering redesigning the oscillator with a relay switched resistor string.
This will solve the problem with the Mdac parasitic capacitance. It will increases the the loop stability and extends the frequency range of the oscillator and the distortion will be lower. In the end using relay is no more expensive that using Mdacs and might even be a bit less expensive.
The multiplier and PI controller will still use Mdacs because this is where they shine.
There is one other possibility which is to use optical current diodes. The diodes outputs a current which can feed a transimpedance amplifier. If the supply to the diode is a variable linear current source then this can be a second input. What I have in mind is the IL300 linear optocoupler. I have one here and intend to do a linearity analysis on it. One might not be enough but a second coupler could be used to feedback to the input of the driver for the optocoupler. Rather than relying on a matched pair of optical diodes.
The ADC in the Shibasoku 725C takes it's input from a gain of 2 buffer which is driven from the final high gain stage.
The monitor port is decoupled from the gain of 2 buffer by an 2:1 2k2 resistor attenuator.
The meter input circuit is taken from the final high gain stage before the gain of 2 buffer amp.
This arrangement improves the SNR of the normal monitor port.
As you know the ADC/DAC has it's own analysis monitor port.
Last edited:
The vp7722A monitor output is from a DAC. It may then go thru a buffer or filter et on its way to the front panel Monitor Output.
8 bits is worse than the opamp s/n so I figure the DAC is the limiting device.
Does the output of the 725 also come from a DAC?
???
-RNM
Last edited:
The normal monitor port is pure analog. It's a normal distortion analyzer.
The analysis monitor output port is off the filters for the analysis. These filters are connected to the DAC. The filters remove the DAC noise, limit the band width to the first 10 harmonics and select the individual harmonics. There is also high pass filter to remove power supply hum and noise. The ADC/DAC are included only to perform the noise reduction and fix the output reconstruction frequency to 488Hz. This simplifies the band pass filters because they can be a fixed frequency. No tuning required for all input frequencies from the DAC.
The Panasonic seems to be built on the same concept as the Shibasoku 725 but simpler.
It appears to be scaled down. If the Panasonic doesn't have a monitor port for the analyzer section, I'll bet you can add one in.
An 8 bit dac will have a poorer dynamic range but in the end might be lower noise.
The reason for this is the quantized steps are larger and the noise won't be flipping bits as much as a higher resolution DAC. It has less sensitivity to the noise. If the signal of interest is amplified, made very large compared to the noise the SNR can be better. It's all relative to the scale.
I have done some mods david wrote on my 725C which I got recently.
Thank you david and thank you demian kindly provide me a 725B manual and schematic.
I was able to draw DC out from data out connector referring the manual.
Replaced A3 QA2 and A4 QA5 to LM4562, yes it is dual opamp, because I don't have suitable single opamp in stock.
Rest one circuit is set to voltage follower and + input is tied to GND.
A7 mod done, by isolating edge finger with polyimide tape instead of cutting trace so that let it restorable.
I did not do other mods of opamps, resistors and A2 but got reasonable result, about -120dB THD+N at 1kHz BW30kHz .
At 10kHz input, 2nd HD is greatly reduced.At 20kHz, residual noise is far larger than 10kHz because frequency range of BEF transitions at 16kHz thus not significant improvement is gained.
I also tried to replace A3 QA1 to singled LM4562 to reduce residual noise further.
I expected replacing it may affect amplitude response upper notch frequency.
The amplitude response did not differ by the mod ,but the opamp weakly oscillate when inputting some specific frequencies such as 997Hz .I don't know why!oh no.
The oscillation occurs after top and bottom portions of swing.
I restored back HA2565 but will try again because A3 QA1 and A4 QA5 affect largely residual noise.
Next I will try to control AG16A with GPIB from PC via arduino and read DC out or VF out from 725C.
My 725C equips optional steep 20k LPF but sometimes got odd result when measuring distortion on delta sigma DAC even with that LPF.
It seemed clipping somewhere in the signal path.
I confirmed the LPF is inserted between A5 and A6 boards so it must be clipping at A4 QA5 x10 amp.
Routing the LPF before A4 or A1 will solve the problem sacrificing residual noise...
Thank you david and thank you demian kindly provide me a 725B manual and schematic.
I was able to draw DC out from data out connector referring the manual.
Replaced A3 QA2 and A4 QA5 to LM4562, yes it is dual opamp, because I don't have suitable single opamp in stock.
Rest one circuit is set to voltage follower and + input is tied to GND.
A7 mod done, by isolating edge finger with polyimide tape instead of cutting trace so that let it restorable.
I did not do other mods of opamps, resistors and A2 but got reasonable result, about -120dB THD+N at 1kHz BW30kHz .
At 10kHz input, 2nd HD is greatly reduced.At 20kHz, residual noise is far larger than 10kHz because frequency range of BEF transitions at 16kHz thus not significant improvement is gained.
I also tried to replace A3 QA1 to singled LM4562 to reduce residual noise further.
I expected replacing it may affect amplitude response upper notch frequency.
The amplitude response did not differ by the mod ,but the opamp weakly oscillate when inputting some specific frequencies such as 997Hz .I don't know why!oh no.
The oscillation occurs after top and bottom portions of swing.
I restored back HA2565 but will try again because A3 QA1 and A4 QA5 affect largely residual noise.
Next I will try to control AG16A with GPIB from PC via arduino and read DC out or VF out from 725C.
My 725C equips optional steep 20k LPF but sometimes got odd result when measuring distortion on delta sigma DAC even with that LPF.
It seemed clipping somewhere in the signal path.
I confirmed the LPF is inserted between A5 and A6 boards so it must be clipping at A4 QA5 x10 amp.
Routing the LPF before A4 or A1 will solve the problem sacrificing residual noise...
Filtering digital is tricky business. Every time a dac switches it triggers the impulse response of the filter. The noise looks like large lobes with distortion products between. I found this out try to measure the output of an IV convertor on a Mdac.
I can post the results here if want to see what it looks like.
I'll look into the oscillation issue you mentioned later today as I just home for lunch.
I can post the results here if want to see what it looks like.
I'll look into the oscillation issue you mentioned later today as I just home for lunch.