Low-distortion Audio-range Oscillator

Does an Mdac have sufficient performance for the low distortion instrument we are seeking here?

What I have here shows that they do. One can also parallel them for better performance.
Mdacs were used throughout the AP SYS1 and no doubt in their newer models but I haven't looked.
The newer Mdacs offer better performance than those from the mid to late 80's.
 
I've completed the modes on the supply. Schematic is shown below. These are the final values. Some measurements.

My Tek main frame has a vertical output. This output is connected to the input of a HP339A set to level mode. 1Vrms is input to the scope and the scope reads 1.414Vp to the best my eyes can see.
The 339A is set to relative adjust and adjusted to read 1Vrms (0dBV).
The 7A22 is set to 10uV/div at 29kHz bandwidth, 1Hz to 30kHz.
The reading on the 339a is at the bottom of the scale so I bumped the input range down -20dBV. With the 7A22 set to the 10uV range, the full scale on the 339A is -100dBV plus the -20dB for the input range adjustment. The total full scale is -120dBV

The positive regulator measures -5.6dBV and the negative reg measure -4dBV
That gives us -125.6 for the pos reg and -124dBV for the neg reg.

Both the positive and negative reg has settled to -126dBv.

With the probe shorted to ground the measurement system noise measures -126dBV.
If the system noise were subtracted from the measurement....
I can't do this unless I know the output impedance of the regulator. To be fair the system noise must be measured with the same impedance terminating the probe.
 

Attachments

  • PSfinal.JPG
    PSfinal.JPG
    99.3 KB · Views: 282
Last edited:
Member
Joined 2004
Paid Member
To get a real handle on the noise of this circuit is more of a project than the circuit itself. I build a transformer coupled preamp with a geoformer with a 35:1 ratio into an LT1115 to get a low noise preamp. The input Z is something like 50 Ohms so its way higher than the output of the regulator. Its noise floor worked out to approx 100 pV so it was below the noise of the regulator. As you discovered the noise is very low.

The figure of merit is the voltage noise for a regulator. Thats what will limit the noise performance of the associated circuitry, if it doesn't have PSRR. The circuit was originally hatched for a low phase noise crystal oscillator. Those do not have alot of power supply noise rejection so a low noise supply is important. however they have no dynamics as a low- essentially a resistor so transient response is not a big issue.

The core "trick" of this regulator is the single junction that compares the output to the reference. That is the lowest noise possible (short of a transformer) to sense and reduce the noise. The second transistor is used as an amplifier. Used as a follower limits the noise to that of the follower. However there is nothing to correct for the transistors thermal response. So put the transistor in a styrofoam box and let the TL431 stabilize the output.

If you mount the transistor socket in the front, behind an access panel, and sell the transistors packed in a styrofoam block, after selecting them for noise you would have the only real "audiophile" fuses.
 
To get a real handle on the noise of this circuit is more of a project than the circuit itself. I build a transformer coupled preamp with a geoformer with a 35:1 ratio into an LT1115 to get a low noise preamp. The input Z is something like 50 Ohms so its way higher than the output of the regulator. Its noise floor worked out to approx 100 pV so it was below the noise of the regulator. As you discovered the noise is very low.

The figure of merit is the voltage noise for a regulator. Thats what will limit the noise performance of the associated circuitry, if it doesn't have PSRR. The circuit was originally hatched for a low phase noise crystal oscillator. Those do not have alot of power supply noise rejection so a low noise supply is important. however they have no dynamics as a low- essentially a resistor so transient response is not a big issue.

The core "trick" of this regulator is the single junction that compares the output to the reference. That is the lowest noise possible (short of a transformer) to sense and reduce the noise. The second transistor is used as an amplifier. Used as a follower limits the noise to that of the follower. However there is nothing to correct for the transistors thermal response. So put the transistor in a styrofoam box and let the TL431 stabilize the output.

If you mount the transistor socket in the front, behind an access panel, and sell the transistors packed in a styrofoam block, after selecting them for noise you would have the only real "audiophile" fuses.

Oh no not just foam. It needs an oven as well. This has to be a very stable fuse for them audiophiles.

We're talking about maybe 40uV of drift and it recovers fast. I don't think this is an issue for op amps. As for the socket. I couldn't find one for the TO92 so I made one from an open frame IC socket. The ones that actually have sockets.

I looked for one of those GEO transformer a while back but couldn't find one anywhere on the net. Have to keep checking ebay I guess. You've mentioned this before.

The measure was at 5mA. At 150mA the ripple pushes the measure to -116dBV. Still respectable. It's darn quiet. Now onto the oscillator. This project is getting really expensive but I'm having fun.
 
What I have here shows that they do. One can also parallel them for better performance. Mdacs were used throughout the AP SYS1 and no doubt in their newer models but I haven't looked. The newer Mdacs offer better performance than those from the mid to late 80's.

Yes, the SYS-2722 uses MDACs for frequency setting and fine level tuning, but those are PCB-level implementations based on discrete resistors and CMOS IC switches.

As far as I understand, the latest revision of it uses a topology based on a R-2R-4R etc. parallel network. The problem with the more common R-2R ladder is that it causes higher noise contribution (e.g. because the noise gain of associated amplifiers is higher). For the System One it was good enough, but as we're approaching -120 dB THD+N (in a 22 kHz BW) things change.

Samuel
 
MDAC-
In a sense even this is a multiplying DAC: http://media.digikey.com/pdf/Data Sheets/NJR PDFs/MUSES72320.pdf It changes the gain based on the input code. For this application how much gain range is needed? Are you tuning with the mdac as well? Is it servo/disciplined to a reference?

I didn't know the muze is now available at Digi. Last time I looked it was just being introduced and it was hard to get. Thanks for the update. It is a candidate for tuning.
 
MDAC-
In a sense even this is a multiplying DAC: http://media.digikey.com/pdf/Data Sheets/NJR PDFs/MUSES72320.pdf It changes the gain based on the input code. For this application how much gain range is needed? Are you tuning with the mdac as well? Is it servo/disciplined to a reference?

Not servo/disciplined, Couldn't do this with a Mdac. Too much disturbance during tuning.
The distortion rises from the dac glitch. But most of the time the osc is not being tuned, just sitting there. Frequency servo would have to be done on an analog basis unless, for the test, the disto doesn't matter. It generates high order harmonics but still at reasonable level.
I didn't think it would work as well as it does.

I'll be building ver2 over the next few weeks so no need to speculate. Lets test it.

Once this ver is done and I'm satisfied I'll ship it down south for Rick and Demian to check out.
 
Last edited:
I've completed the modes on the supply. Schematic is shown below.

My biggest complaint is the way you have drawn R14. It has a dot to show where it is connected, but this is still very close to the way you show other lines crossing but not connected. if you moved the part one grid to the right it would be much harder to mistake that it is connected not jumping the line.

I suspect the power supply can be made at least 15 -20 db quieter!

You have a series RC across the power transformer. A parallel form works better. The capacitance should also rise to about 1-10 uf. The way I size the R is to load the transformer to 10% of capacity, that is where the magnetic coupling of the transformer is very dominant compared to the capacitive leakage. The the capacitor value is chosen to minimize transformer resonance on the secondary with the primary shorted.

I also place small capacitors across the diode bride. For a cheap diode bridge 220 pf is a good choice. For a low switching noise bridge, 47 pf. This will pass more line noise than the diodes by themselves but will eliminate switching spikes at 500 - 1000 khz.

The next improvement would be to use two bridges and not connect the transformer center tap to ground. By creating the ground or common after the diode bridge and filter capacitors you get 20 db less RF leakage to the ground. ( A transformer does pass line noise, a diode when not conducting power does not!) Better still with the two bridges is to wire one of them out of phase with the other. That way any common mode line noise becomes differential and can be cancelled. Of course a good RF shut on the output of each bridge is useful. I would use an NPO (also called COG) ceramic capacitor of 100 pF here.

I would also use 5 x 1000 uf filter capacitors rather than a single 4700 uf. Even better would be 2000 uf 1R 2000 uf 1R 1000 uf.

Now when you look at the output spectrum if you see power line harmonics the structure is important. If you see even and odd order harmonics that is actually OK. If you see just odds or they are dominant it means either your transformer is not properly shielded or your grounding scheme isn't working. If you see dominant evens that is from radiated energy and can often be cleaned up by a well placed capacitor. If you don't see any harmonics, you are just not looking closely enough.

Now if you want to really go to extremes use a thermo-electric cooler on the output driver transistor!

ES
 
Last edited:
My biggest complaint is the way you have drawn R14. It has a dot to show where it is connected, but this is still very close to the way you show other lines crossing but not connected. if you moved the part one grid to the right it would be much harder to mistake that it is connected not jumping the line.

I suspect the power supply can be made at least 15 -20 db quieter!

You have a series RC across the power transformer. A parallel form works better. The capacitance should also rise to about 1-10 uf. The way I size the R is to load the transformer to 10% of capacity, that is where the magnetic coupling of the transformer is very dominant compared to the capacitive leakage. The the capacitor value is chosen to minimize transformer resonance on the secondary with the primary shorted.

I also place small capacitors across the diode bride. For a cheap diode bridge 220 pf is a good choice. For a low switching noise bridge, 47 pf. This will pass more line noise than the diodes by themselves but will eliminate switching spikes at 500 - 1000 khz.

The next improvement would be to use two bridges and not connect the transformer center tap to ground. By creating the ground or common after the diode bridge and filter capacitors you get 20 db less RF leakage to the ground. ( A transformer does pass line noise, a diode when not conducting power does not!) Better still with the two bridges is to wire one of them out of phase with the other. That way any common mode line noise becomes differential and can be cancelled. Of course a good RF shut on the output of each bridge is useful. I would use an NPO (also called COG) ceramic capacitor of 100 pF here.

I would also use 5 x 1000 uf filter capacitors rather than a single 4700 uf. Even better would be 2000 uf 1R 2000 uf 1R 1000 uf.

Now when you look at the output spectrum if you see power line harmonics the structure is important. If you see even and odd order harmonics that is actually OK. If you see just odds or they are dominant it means either your transformer is not properly shielded or your grounding scheme isn't working. If you see dominant evens that is from radiated energy and can often be cleaned up by a well placed capacitor. If you don't see any harmonics, you are just not looking closely enough.

Now if you want to really go to extremes use a thermo-electric cooler on the output driver transistor!

ES

Thanks Ed. I'll take all this into consideration. I'm out of board real estate as it but I can make the board bigger if necessary. Question is does it need to perform better than it is. I was originally using two bridges but the board was getting tight for space and they charge by the square inch.

I'll fix the hard read in the schematic.
 

Attachments

  • PSfinal.JPG
    PSfinal.JPG
    103 KB · Views: 299
Last edited:
The MDAC in S1 was a custom AD device; customised in the sense that the (laser trimmed) feedback resistor was inplemented on-chip and thus tracked the MDAC resistors with temperature.
There's a patent by Bruce Hofer on it from early 80-ies or so.

Jan


The AD5543 and others have an internal laser trimmed FB resistor. I guess AD liked the idea and incorporated into their new designs.
 
Thanks Ed. I'll take all this into consideration. I'm out of board real estate as it but I can make the board bigger if necessary. Question is does it need to perform better than it is. I was originally using two bridges but the board was getting tight for space and they charge by the square inch.

I'll fix the hard read in the schematic.

I would not connect a transformer center tap to the ground. Two bridges are a big improvement. If you don't want that use two inverse parallel connected diodes in the ground lead. The diodes only conduct when current flows, so you get as much as 10 db lower line noise!
 
I would not connect a transformer center tap to the ground. Two bridges are a big improvement. If you don't want that use two inverse parallel connected diodes in the ground lead. The diodes only conduct when current flows, so you get as much as 10 db lower line noise!

I have room on the board for two bridges. It's a small change to the layout. The RF filter components are small enough and easily fit. But I don't have room for RC mains filtering unless I enlarge the board. Peltier cooler might be a bit extreme.

These transformers put out about a 5% clipped sine which might complicate the spectrum analysis you described.

"The diodes only conduct when current flows, so you get as much as 10 db lower line noise"

Then the ground is floating most of the time. Interesting.
 
Last edited: