Accuracy when Using a Comparator / OpAmp to Find Peak Voltage of CW Signals

[niconiconi]

I need to measure the peak voltage from a CW signal generator as a check of its frequency flatness (if anyone saw my thread on another board of this fourm, for building a leveled sinewave calibrator). Output level is 0.5 Vpp to 5.5 Vpp into 50 Ω, frequency is from 0 to 50 MHz. Because the signal generator itself uses peak detection for leveling, I need a different way to verify its operation and to ensure measurement confidence.

The oscilloscope itself cannot be used - the CW signal generator itself is an oscilloscope calibrator... . Also, I want to avoid any instrument that responds to RMS, for oscilloscope calibration, one needs peak voltage, even a harmonic distortion as low as 40 dBc can create a Vrms to Vpp conversion error around 1% in the worst case. It means there's not much choice left for me...

For frequencies above 50 MHz, I can use a calibrated peak-responding RF power meter, but it's unable to verify the flatness under 50 MHz (technically some meters are usable to 10 MHz, though, I still wish to have an overlapping measurement range up to 50 MHz for better confidence). After some thoughts, I was inspired by the vintage Tektronix 7A13 "differential comparator", thus I'm wondering whether it's possible to use a comparator or opamp for this task - connect the DC reference to one input and connect the CW signal to another input, the output should start toggling. One adjusts the reference until the output flatlines and stops toggling, then measures the reference on a DMM as the peak voltage.

(BTW, Another alternative idea is building a sampling-and-hold circuit and doing equivalent time sampling on a DMM).

The question is, how much accuracy should I expect from the comparator approach? Is it possible to achieve an accuracy better than 1%?

At the low end of 0.5 Vpp or 0.25 Vp, 1% is 2.5 mV, so I expect DC offset and noise to be a major problem. The DC offset can probably be calibrated out, but the noise, I'm not so sure, especially at 50 MHz. But does it have at least a chance to work at 5 Vpp?

[Terry Bites]

Do you really need to go to "0 MHz" ?
Log amps are easy to use and solve all those dynamic range problems They're not cheap but they do just work with no faff,
I've been a long time fan of the AD8307 (sad).  Similar but faster is the AD8310. I think I'd be looking at the lower cost modern part LT5537 now.
The ADL5511 env+rms looks FAB but a is bit more pricey. I use the decent samwich counting method of evaluating VFM.
Loads of pre built AD8307, AD8310, modules on ebay- who knows?

[niconiconi]

Do you really need to go to "0 MHz" ?
Log amps are easy to use and solve all those dynamic range problems They're not cheap but they do just work with no faff,
I've been a long time fan of the AD8307 (sad).  Similar but faster is the AD8310. I think I'd be looking at the lower cost modern part LT5537 now.
The ADL5511 env+rms looks FAB but a is bit more pricey. I use the decent samwich counting method of evaluating VFM.
Loads of pre built AD8307, AD8310, modules on ebay- who knows?

Sorry, but this doesn't answer my question. The original question is, whether it is possible to verify the CW peak voltage flatness from 0 to 10 MHz (better to 50 MHz) with high accuracy, inexpensively with some purpose-made test circuits (not some unobtanium, NIST-traceable, in-house test fixtures used by Tektronix with an astronomical price tag), since peak-responding RF power meters don't work in this frequency range.

Speaking more about log amps and my architecture - which is off-topic but I have to point out the background to avoid future misunderstandings...

My current prototype already uses a Analog Devices log amp for leveling, it's the AD8361, a RMS-responding power detector (It's not exactly a proper log amp, since its output is linear, not linear-in-dB, but it's almost one since the architecture is similar, and it also has a closely related cousin, the AD8362, which uses a logarithmic output and is a true log amp). It works down to near-DC, and my quick test (using an actual HP microwave power sensor, using the direct comparison method), shows its flatness is excellent, around 0.3 dB from 50 MHz to 1 GHz, and around 0.1 dB from 50 MHz to 200 MHz.

Unfortunately, now I have two problems.

1. The AD8361 is a RMS-responding power detector, which is great and what you usually want for RF measurements. However, for oscilloscope calibrators, what needed is an accurate peak voltage, not an RMS voltage. In fact, a lot of harmonic distortions can be tolerated as long as the peak voltage is accurate. On the other hand, if you use RMS voltage to control the peak voltage, now you must ensures low distortion, otherwise it creates a significant error in the worst case. Thus, I'm planning to scrap my current AD8361 prototype and redesign it with an "envelope power detector" type. AD provide many options as well, they respond to peak rather than RMS. Thus I'm referring them categorically as "peak detectors". The ADL5511 is exactly what I'm considering.

2. The typical flatness of a log amp (and also the design goal of this calibrator) is around 0.3 dB, or 3%-4% in amplitude, which is also exactly what I need for the calibrator. But you see, to verify its flatness, an instrument with higher accuracy must be used. So you can't use any log amp for this verification, it needs to be something completely different. For 10 MHz and above, my solution is a microwave power sensor. But a solution is still needed under 10 MHz. I specifically mentioned 1% amplitude accuracy as a performance target, since it's around 0.1 dB, thus better than the 0.3 dB device I'm trying to verify.

Long story short, I need a test circuit with an accuracy better than the ADL5511 in order to characterize the ADL5511 itself! Or at least, an accuracy comparable to ADL5511, but using a different theory of operation, thus I can get some cross-validation and measurement confidence.

Thus the original question and the two options I'm considering: comparators, or sampling-and-hold plus equivalent time sampling.

[niconiconi]

Thus the original question and the two options I'm considering: comparators, or sampling-and-hold plus equivalent time sampling.

I just found the "sampling-and-hold" plan I was describing is my personal reinvention of a sampling voltmeter. after seeing the following...

the HP 3406A uses a unique "incoherent" sampling technique to translate a high frequency input signal into a low frequency equivalent [...] The Sample Hold output provides a statistically equivalent signal to the input which gives true rms, peak voltage, amplitude modulation envelopes and pulse height information.

This is exactly what I was considering to build. Now... I think I just need to find a HP 3406A at the flea market and all my problems can be solved.

[Terry Bites]

But thats no bad thing. Shine some modern light on an old idea. 

[niconiconi]

The oscilloscope itself cannot be used - the CW signal generator itself is an oscilloscope calibrator... . Also, I want to avoid any instrument that responds to RMS, for oscilloscope calibration, one needs peak voltage, even a harmonic distortion as low as 40 dBc can create a Vrms to Vpp conversion error around 1% in the worst case. It means there's not much choice left for me...

I just realized I was overconstraining myself, the actual solution should be simple. Both the oscilloscope and a RMS responding meter can be used. First, use a signal generator and a wideband RMS voltmeter to characterize the frequency response of the oscilloscope, then the "calibrated" oscilloscope is used as a transfer standard to measure the peak-to-peak voltage of the signal generator.

This assumes the Vrms and Vpk measured by the oscilloscope has almost the same frequency response and error, which is a reasonable assumption.