Distortion Measurement

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

mike-wsm

Member
Joined
Mar 27, 2017
Messages
24
Location
Weston super Mare UK
Hello guys,
I was at Radford Electronics during the mid-seventies as a design engineer working on LDO and DMS. The technology seems to have faded away so I am currently designing and building a High Purity Sine Wave Generator (LDO) and two types of Distortion Meter (DMS), a stand-alone twin tee type and a subtractive type that has to be linked to the oscillator.
The units are similar to the Radford ones except that I am avoiding the use of expensive custom pots and switches. They should provide similar if not identical performance.
They are currently in build with circuit diagrams shown at http://mike-wsm.org.uk/g3.html
I hope to have them all working by mid-2017. Bear with me, old age takes its toll.
Enjoy!
mike
 
mike-wsm said:
Hello guys,
I was at Radford Electronics during the mid-seventies as a design engineer working on LDO and DMS. The technology seems to have faded away so I am currently designing and building a High Purity Sine Wave Generator (LDO) and two types of Distortion Meter (DMS), a stand-alone twin tee type and a subtractive type that has to be linked to the oscillator.
The units are similar to the Radford ones except that I am avoiding the use of expensive custom pots and switches. They should provide similar if not identical performance.
They are currently in build with circuit diagrams shown at http://mike-wsm.org.uk/g3.html
I hope to have them all working by mid-2017. Bear with me, old age takes its toll.
Enjoy!
mike
Thanks for sharing and keep us informed.  Getting old isn't for wimps.

I don't think I've seen many OTAs used in AGC loops but no reason why not.  What THD level are you targeting?

You mention quadrature in your level pick-off but I only see one phase being used. There is an old trig trick, to square two phases and sum them  to realize trig identity (Sin^2+Cos ^2=1). I don't know that I've seen a successful realization of this but we did punt it around here years ago. Most just smooth the AGC loop, sometimes with adaptive time constants for quicker settling.

One quick question, the op amp in the top middle of the low distortion sine wave circuit, appears to have a capacitor in the negative feedback path, I do not see a DC negative feedback path for that one op amp through the associated circuitry.

wrt the subtractive distortion analyzer you say that you subtract I and Q (quadrature) outputs which appear to be 90' apart, to reject the fundamental. In my experience 90' phase shifted signals does not null fully. 

But I am still on my first cup of coffee so may be mistaken,.  Welcome to the forum.

JR
 
Hello John,

Thanks for your comments.

LDO: The op amp top middle is the control integrator. It responds to the sampled amplitude and controls the negative resistance. So yes, there is in effect a dc path all the way round the system.

The quadrature squared edge is used to generate the sample pulse which has to be at the peak of the reference phase.

Yes, the sum of squares technique is one I considered way back in time and it was later used in an app note by Analog Devices or Burr Brown. It is not guaranteed to be ripple-free because of tolerances. Sample-hold when used on one peak is the same on every cycle.

The ultimate resolution of the Radford instruments turned out to be 1ppm or 0.0001 per cent at 1kHz. I am hoping to get 0.001 percent using off-the-shelf components first time round, perhaps better in the next generation.

DMS: By adding appropriate proportions of +/-I and+/- Q you can generate any desired vector. So yes, it does null exactly. I have not added autonull yet, this is a treat in store for you, it used four OTAs to detect, feed back and reinject I and Q. In those days we used CA3080.

Thanks for your kind welcome, nice to be here,
mike
 
mike-wsm said:
Hello John,

Thanks for your comments.

LDO: The op amp top middle is the control integrator. It responds to the sampled amplitude and controls the negative resistance. So yes, there is in effect a dc path all the way round the system.
Thanks, that explains it.
The quadrature squared edge is used to generate the sample pulse which has to be at the peak of the reference phase.
Got it...  I think Sam Groner did some sampling window tricks in his low distortion oscillator.

http://www.nanovolt.ch/resources/low_distortion_oscillators/

I suspect you are familiar with Robert Cordells Osc/analyzer project published in Audio Magazine years (decades) ago.
Yes, the sum of squares technique is one I considered way back in time and it was later used in an app note by Analog Devices or Burr Brown. It is not guaranteed to be ripple-free because of tolerances. Sample-hold when used on one peak is the same on every cycle.
yup...
The ultimate resolution of the Radford instruments turned out to be 1ppm or 0.0001 per cent at 1kHz. I am hoping to get 0.001 percent using off-the-shelf components first time round, perhaps better in the next generation.
nice... after I went over to the dark side (digital) it is hard to ignore just banging bits into a precisions DAC.
DMS: By adding appropriate proportions of +/-I and+/- Q you can generate any desired vector. So yes, it does null exactly. I have not added autonull yet, this is a treat in store for you, it used four OTAs to detect, feed back and reinject I and Q. In those days we used CA3080.
Yup the 3080 is an old work horse... Did you ever mess with the 3280.? supposed to be lower noise with extra linearized input components. I used it in an audio kit back in the 80's
Thanks for your kind welcome, nice to be here,
mike
We can always use another old analog dog (boffin)...

JR
 
wishing you success.
devices were employed at Neve, performed very well.
hope you come up with a noise free, battery eliminator.
 
> In those days we used CA3080.

CA3080 was out of production a long time.

(Apparently in 2005 Intersil gathered-up their in-production tooling and moved to Florida. The '3080 had not been run in a while, that tooling was lost when they sold the building.)

The '3080 are in small demand for guitar pedals etc. The market is full of fakes.

Small Bear Electronics and Rochester Electronics worked together to get a run of new '3080, apparently on the old masks, on an old process. You can buy from either. SBE's price is a bit better.

The LM13700/13600 is the same idea, with a 2-D Iabc port, and a buffer thrown in free. This *appears* to be in good supply. Mouser shows the through-hole as 800 in stock at $2/each. SOIC-16, 2+K stock $1/pop. DigiKey similar. Thru-hole marked "Not For New Designs", most SOIC are "Active".

Might be worth planning for either/or, to accommodate long-term supply changes.
 
gridcurrent said:
wishing you success.
devices were employed at Neve, performed very well.
hope you come up with a noise free, battery eliminator.
Funny you should mention that, just working on a LT1553 low noise SMPS, might have a pcb milled by the weekend. I havent got the correct transformer yet, but a Wurth Flex is at hand for the proto
 
Martin Griffith said:
Funny you should mention that, just working on a LT1553 low noise SMPS, might have a pcb milled by the weekend. I havent got the correct transformer yet, but a Wurth Flex is at hand for the proto

1553 or 1533? The latter is a pretty great switcher.
 
Thanks for the battery eliminator advice. I intend use PP3 batteries with 78L05 and 79L05. The Kodak alkalines at our local store are one pound (dollar) each and provide 300mAh, should last a good while. This fits with my kitchen table range of battery powered instruments.

For a mains powered version of the LDO I would go for 78L15 and 79L15, there is so little current to supply and yes smps noise is to be avoided. The higher voltage would permit more signal for the same noise, and higher output voltage. A better output attenuator would be nice but more switch wafers are needed for a good design, we needed separate wafers for the high and low signal sections to prevent feed-through. Perhaps two separate switches, um.

If the osc is mains powered then the DMS of either type needs to be battery powered to avoid hum loops. I intended to use built-in panel meters but available LCD types are 7106 derived and need a floating supply, so my display is in a separate enclosure. Yet another PP3.
 
Hurrah! Looking at the professional quality attenuator in the Radford LDO with its strange resistor values and multiple wafers I found I could scale the values to make one with preferred values at a slightly lower impedance, and then add a single padding resistor at the output to gain the required output impedance. I managed to do this for audio 600 ohm and 150 ohm attenuators and also for a radio frequency 50 ohm attenuator, using mostly E6 resistors.

There are significant advantages in doing it this way, only one switch wafer is required for 0 to -100dB in 10dB steps and there are only 20 resistors to mount on the switch, 10 around the periphery and 10 radial to a central ground star, plus one input resistor and one output padding resistor, making fabrication an easy task. The twelfth switch way has to be grounded to prevent capacitive coupling from the highest signal level to the lowest.
 
Dem weh de days ... when the Radford LDO & DMS was the bee's knees.

You didn't by any chance know a Mr. Wallis from those days?  His brother Peter Wallis worked with me at Wharfedale for many years.
 
Sorry, don't think I met Mr Wallis during my two spells there during the mid and late seventies. Many engineers moved through Radford, leaving their mark with a new generation of audio amplifiers or test equipment, before moving on and starting up their own companies. I shall soon be meeting with a former colleague who started as a Radford apprentice and was with Radford much longer, I'll ask him.
 
Re obsolescence of OTA type devices, our brilliant Danish engineer, noted for his brief feat of self-aviation assisted by certain substances, used a gain control element from Japan which was an LDR (light dependent resistor) potted with a LED. The LDR was remarkably linear. and at any one light level and provided extremely low distortion. I suppose it is possible to make a replica by fixing an LDR and a LED together in an opaque tube. The LDR could be wired from the inverting amplifier of a biquad back to the input of the second integrator to provide the desired negative resistance effect.
 
Such thingy is called "Vactrol", and is sadly now banned for commercial use across the EU (because of the minute amounts of Cd in the photocell). You can still buy cheap Chinese knockoffs, search for part number e.g. "M1210CLC" or "VTL5C" on aliexpress..

Jakob E.
 
Thanks! Found both devices available within EU in RoHS versions. Response times are asymmetric, 4ms on and 50 ms off (to 100k) for the VLT5C9, not too bad. Shall have to consider doing a Vactrol version of my sig gen.
 
Trying to avoid obsolete technology If price is no object, modern VCAs (like from THAT corp) are quite linear but perhaps overkill for oscillator loop AGC. (and VCAs may go away in a few decades too).  :'(  (nah- too expensive)

The digital devil sitting on my left shoulder is suggesting using a DPOT while that could be scaled for sensitivity and a simple up down control circuit could be managed in analog glue, it seems it would always be hunting up/down.  (nah- too digital)

The popular approach of a JFET linearized by adding some AC drain voltage back into the gate -6dB, seems respectable enough.  The topology Robert Cordell used in his published oscillator from a few decades ago looked pretty clever IMO.  ( Page 39  http://www.cordellaudio.com/papers/thd_analyzer.pdf. ) You could substitute modern op amps and lower noise JFETs. That don't appear to be going away anytime soon.  You probably don't need to invest in his complex FW rectifier, that as you note could be a source of error (Cordel added a symmetry trim there, but didn't tweak distortion in the JFET, sampling seems less prone to ripple). 

JR

PS: Sam rolled his own multiplier for AGC gain control. (http://www.nanovolt.ch/resources/low_distortion_oscillators/). It's hard to imagine eclipsing Sam's design.
 
Er, yes, used a linearised JFET (add half the drain ac to the gate) in a Wien bridge oscillator but I'd take a lot of convincing that it was anywhere near the 60dB s/(n+d) of a good old CA3080.

There was a grand announcement from University of Bath in its younger days that they had invented a marvellous new floating input voltage-to-current amplifying device which sounded remarkably like a 3080....

.....then they went very, very quiet about it.
 
mike-wsm said:
Er, yes, used a linearised JFET (add half the drain ac to the gate) in a Wien bridge oscillator but I'd take a lot of convincing that it was anywhere near the 60dB s/(n+d) of a good old CA3080.
The late Jim Williams would disagree. In Linear Technology Application Note 43 (page 28ff) he shows a JFET-stabilised Wien bridge oscillator with a measured distortion of 0.0018%, or better than 90dB.

(He does admittedly go on to point out that the JFET needs a device-dependent trim, and produces a vactrol-based design which, with some other improvements, yields a further 15dB improvement in THD. One of these days I need to find the time to apply these improvements to a JFET-based design, to compare apples with apples.)

JDB.
 
Yes, but that is a highly tweaked circuit where the FET is contributing as little as possible to the overall circuit. The remaining distortion is a rather nasty combination of integrator ramp and FET second harmonic contribution. You would be hard pressed to distinguish artefacts introduced by amplifiers under test.

edit - Perhaps I should explain that my figure of 60dB refers to the CA3080 itself. When used appropriately in an oscillator the THD is less than 1ppm or -120dB. Your figure of -90dB relates to the complete circuit, not to the FET itself. There is also a difference in the content of the distortion, the FET is asymmetric and produces even harmonics, whereas the CA3080 is balanced and produces only odd harmonics.

Overall one needs to make an instrument which performs well at all frequencies and for all users. The biggest problem is 'bounce' where any adjustment of frequency causes the amplitude to bounce until it finally settles. This is especially bad at low frequencies, say 10Hz. Early Radford LDOs had a switch to select fast settling or low distortion. Later ones with the s/h method of tracking amplitude up and down are completely bounce-free, you can tune them as fast as you like and they settle instantly.
 

Latest posts

Back
Top