Pls recommend a signal generator with external pll input

[Bxaresearch]

Hello,

I am looking for a signal generator or clock synthesizer that would allow me to double, triple etc an external sine wave signal frequency and produce low phase noise square wave locked with high precision to input 1khz frequency. If I feed 1khz 1V, I need to obtain 2khz, 3khz etc that are same phase as input.
Using a multi output generator that provides signals locked in phase out of the same instrument is not an acceptable solution.
I would prefer a bench top instrument that can be calibrated in order to achieve certified measurements

Thanks!

[jbb]

It seems like an odd function for a bench top instrument.  Maaaaaybe some audio analyser or lock-in amplifier would offer it as an optional output?

Do you need 2kHz & 3kHz & n kHz outputs at the same time? Or is a single output that you can switch OK?

If you're forced to make your own... it might be a bit tricky because that 1 kHz input freq is pretty low? 

As I understand it, PLLs work best when the Phase Detector operates at a high frequency (many chips like 1 - 100 MHz).  With a 1 kHz input, the Phase Detector can only deliver 2,000 updates per second.  During the 500 us between control updates the analog voltage control signal to the Voltage Controlled Oscillator would wander around a bit and make some phase noise.  And making precision analog electronics for low frequencies can be a bit tricky.

Maybe some digital PLL would help; digital circuits are really good at holding a constant control signal over long periods.

There are clock synchroniser chips which are designed to work with really low input frequencies like 1 pulse-per-second from GPS receivers.  The [a href=https://www.ti.com/product/LMK5B12204] LMK5B12204[/a] from Texas Instruments (never used it, expect there are similar ones for less $$ about) uses a crystal oscillator for short term stability and has a fancy digital PLL inside which can lock onto a >= 1 Hz reference.  It could generate a 10 MHz output locked to the 1 kHz input signal.  But doesn't seem to offer phase synchronisation...

[Bxaresearch]

Jbb, thanks for the reply, you made some good suggestions I will ponder on. As you can imagine I decided to post the question after some agonizing :-) Let me delay the exact reasons behind the requirement but you have hinted on some of them. I would be happy with just the 2x frequency and 3x frequency etc, not all at the same time

[jbb]

You're welcome.

This one bounced around my head a bit, so I've come for a second bite.  Setting aside the phase lock to incoming 1 kHz signal, there are some requirements to get coherent harmonics out...

• To operate PLL, require f_osc is a multiple of 1 kHz
• To get 2 kHz output, require f_osc is a multiple of 2 kHz
• To get 3 kHz output, require f_osc is a multiple of 3 kHz
• Thus to have the choice between (or multiple outputs) 2 and 3 kHz, f_osc must be a multiple of both 2 kHz and 3 kHz, i.e. a multiple of 6 kHz
  

As the number of harmonics of interest goes up, so too does the required f_osc.

Let's imagine we'd like the option to output the following: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 kHz
We can decompose those to products of primes: 1, 2, 3, 2², 5, 2*3, 7, 2³, 3², 2*5
Thus we could use f_osc = product(2³, 3², 5, 7) = 2520 kHz

Let's imagine we'd like to add on 11, 12 kHz
That's going to add some more products of primes: 11, 2²*3
Thus we could use f_osc = product(2³, 3², 5, 7, 11) = 27,720 kHz

And if we want to go a bit further and add on 13, 14, 15kHz
That's going to add some more products of primes: 13, 2*7, 3*5
Thus we could use f_osc = product(2³, 3², 5, 7, 11, 13) = 360,360 kHz

So if we want to use a single value of f_{osc[\sub] and output many different harmonics then fosc needs to be quite high.  Adjusting fosc ratio to vary depending on which harmonics we want to output could make a huge impact, though}

[Alex Nikitin]

You could synchronise 1kHz input with, say, 12kHz oscillator and than divide it by 6 and by 4. That will give you synchronised 2 and 3kHz square wave outputs. The advantage however is that you don’t need to wait for the outputs to synchronise if you change the frequency.

Cheers

Alex

[Bxaresearch]

Jbb, I did not expect.to find the prime numbers here, but I am following that. I see that with a single frequency reference I can achieve multiple integer ratios, interesting technique.
Also Alex reminded me one other possible and perhaps obvious solution that I can't use: synchronizing the signal generator to an external 10Mhz reference. The generators I have are ultra low distortion ones from R&S and Stanford research. I have the UPV,  DS360 as well as SR1. They do have an external clock input, but as I confirmed with manufacturers it is not phase locked  to the signal. I learnt it the hard way also.  It did sound odd to me as it defeats the purpose.
So from what I see, my remaining option is to generate a separate clock and yes, I realize difficulty of that. I was hoping that there might be some device that I'm not aware of that does that. After all, I see chips like ZL30253 but I'm not sure if it will work and if it does, I'll have to implement everything from scratch. But that's the only option I see that can even remotely approach the requirement.
Of course I would prefer a desktop box that is made and calibrated to make things much easier

[Alex Nikitin]

There is another way of doing it and provide a low distortion time correlated output (or several) - a PC with a decent sound card and a sound editor. Create a file containing three (or more) channels and output the lot simultaneously, you would only need to switch channels.

Cheers

Alex

[Bxaresearch]

Yes the sound card has been tried. The issue with sound cards is uncertifiability of the measurement

[Alex Nikitin]

  Nothing stops you from using a certified AC voltmeter and a frequency counter to monitor the signal, log the results and guarantee the accuracy. My guess is that an arbitrary function generator is not good enough distortion-wise for your application?

Cheers

Alex

[jbb]

...
The generators I have are ultra low distortion ones from R&S and Stanford research. I have the UPV,  DS360 as well as SR1. They do have an external clock input, but as I confirmed with manufacturers it is not phase locked  to the signal. I learnt it the hard way also. 
...

All right! So, you're generating the 1 kHz sine wave (I was thinking that you had some 1 kHz signal coming in from the wild).

Can you get a two-output function generator with sufficient capabilities?

[Bxaresearch]

The DS360 is the lowest distortion two signal function generator available, but two channels are not phase locked. I have explored this avenue. The 1khz is pretty much coming from the wild because it goes via DUT at which point whatever the phase relationship there was with the second channel will be compromised.
It turns out that even the manufacturers of the above mentioned devices have no metrology method to certify the distortion level of their devices. Pretty much what is certified and traceable is just frequency and voltage.
I think I might have a certification procedure for thd that is based of frequency and voltage standards but it hinges on being able to generate the 2x 3x etc clock signal that's phase locked to fundamental frequency

[Alex Nikitin]

Keithley 2015 has distortion measurements accuracy specified.

Cheers

Alex

[Bxaresearch]

Thanks for Keithley recommendation, Alex. I remember looking at it in the past while selecting generators. I have not talked to the tech or engineering department there so I don't know what their calibration procedure is. Anybody has one and sent it to calibration? What devices are specified on the calibration sheet for Keithley 2015?

[Alex Nikitin]

Keithley 2015 Service Manual.

Cheers

Alex

[Bxaresearch]

Alex, thanks, I had a look at section 2.7 in the Keithley manual. They specify the DS360 as the distortion calibration standard. The DS360 is calibrated using SR1. The SR1 in turn is not calibrated in its distortion measurement capabilities. That I know first hand. So my original quest is to effectively come up with THD calibration procedure

[jbb]

... The 1khz is pretty much coming from the wild because it goes via DUT at which point whatever the phase relationship there was with the second channel will be compromised. ...

... So my original quest is to effectively come up with THD calibration procedure

Hmm.  I can see why you'd like a very clean sine wave input, then.  Does that nifty Sanford Research box have a 10 MHz clock input or output handy?  Also, the thought occurs that the signal generators may not hit 1 kHz exactly, but instead do some rounding...

It seems like at this point the options are:

• build a Phase Locked Loop (not just Frequency Locked) to get precise phase alignment
• use a shared 10 MHz reference clock and compensate for the circuit under test's phase delay
• build a Frequency Locked Loop and compensate for the circuit under test's phase delay
• pursue a different measurement strategy

For 2. and 3.:

• a matched f_osc clock frequency would provide a stable (but unknown) phase relationship
• could you set your f_osc divider to 1 kHz and measure the phase of the incoming signal?
• could you then calculate phase for 2 kHz, 3 kHz etc?

For 4.:
It seems like getting nice phase alignment will cost you some time & trouble.  Perhaps at this point you should re-visit your goals for the system; do you really need that phase information, or will magnitude information be acceptable?

If you're interested in a THD measurement, would it be adequate to deploy an analog filter (with high Q factor components) that reduces the fundamental amplitude enough that the harmonics are measurable with standard equipment?  That's what's shown in the DS360 manual in the Performance Tests section.

[Bxaresearch]

Let me take this in stages, as it had been 3 years since I started looking at this problem.
So my original question is I believe what I need, weather there is an ready made device.or if I have to make it
The 4th order LC elliptical filter and incredible amount of EM shielding is already involved in rejecting the fundamental
I will continue to comment

[Bxaresearch]

Basically what is needed and does not exist anywhere is low distortion generator that can lock on external 10MHz reference or generator that can provide its internal 10MHz reference so that another generator can lock its 10MHz clock on and generate 1kHz locked to that reference. Ether way the locked phase relationship is the key to the measurement. Short of that, anybody that claims THD figures of their devices is kinda safe if it is in the -80dbc neighborhood (KEITHLEY) but beyond that, there is no guarantees other than the progress that has been achieved by modern 24bit dac/adc topologies and that hinges largely on spice simulations.
What is easier:
1. Generating 10MHz clock that is phase locked to external 1khz
2. Dual output function generator which guarantees phase lock but has poor distortion that has to be improved from -80dbc to -130dbc

[Bxaresearch]

So I was hoping that generating 2kHz square wave given a very stable and digitally generated 1kHz sine out of DS360 would be the easiest

[srb1954]

Basically what is needed and does not exist anywhere is low distortion generator that can lock on external 10MHz reference or generator that can provide its internal 10MHz reference so that another generator can lock its 10MHz clock on and generate 1kHz locked to that reference. Ether way the locked phase relationship is the key to the measurement. Short of that, anybody that claims THD figures of their devices is kinda safe if it is in the -80dbc neighborhood (KEITHLEY) but beyond that, there is no guarantees other than the progress that has been achieved by modern 24bit dac/adc topologies and that hinges largely on spice simulations.
What is easier:
1. Generating 10MHz clock that is phase locked to external 1khz
2. Dual output function generator which guarantees phase lock but has poor distortion that has to be improved from -80dbc to -130dbc

What do you classify as low distortion?

An set of HP 8904A opt 005 synthesisers could generate multiple phase locked signals but the signal purity at -66dBc might not be sufficient for your purposes.

You may need to add some external filtering to clean up the signals but that will potentially alter the exact phase relationships of the signals. However, the phase offsets in the individual 8904As could be adjusted to compensate for any such phase errors.

[srb1954]

You're welcome.

This one bounced around my head a bit, so I've come for a second bite.  Setting aside the phase lock to incoming 1 kHz signal, there are some requirements to get coherent harmonics out...

• To operate PLL, require f_osc is a multiple of 1 kHz
• To get 2 kHz output, require f_osc is a multiple of 2 kHz
• To get 3 kHz output, require f_osc is a multiple of 3 kHz
• Thus to have the choice between (or multiple outputs) 2 and 3 kHz, f_osc must be a multiple of both 2 kHz and 3 kHz, i.e. a multiple of 6 kHz
  

As the number of harmonics of interest goes up, so too does the required f_osc.

Let's imagine we'd like the option to output the following: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 kHz
We can decompose those to products of primes: 1, 2, 3, 2², 5, 2*3, 7, 2³, 3², 2*5
Thus we could use f_osc = product(2³, 3², 5, 7) = 2520 kHz

Let's imagine we'd like to add on 11, 12 kHz
That's going to add some more products of primes: 11, 2²*3
Thus we could use f_osc = product(2³, 3², 5, 7, 11) = 27,720 kHz

And if we want to go a bit further and add on 13, 14, 15kHz
That's going to add some more products of primes: 13, 2*7, 3*5
Thus we could use f_osc = product(2³, 3², 5, 7, 11, 13) = 360,360 kHz

So if we want to use a single value of f_{osc[\sub] and output many different harmonics then fosc needs to be quite high.  Adjusting fosc ratio to vary depending on which harmonics we want to output could make a huge impact, though}

As the input clock frequency becomes impracticably high for generating the higher harmonics another approach might be to start out with a 1kHz train of narrow pulses. This will contain a high level of harmonics and each desired individual harmonic could be then picked out using a suitable narrow band-pass filter.

[ch_scr]

If one is satisfied with fixed 1kHz and multiples, maybe an (series of) injection locked Wien bridges might be the way to go? There are some published designs with extremely low distortion at -140dB. For the synchronized injection locking, the harmonic content of a 1kHz pulse as mentioned above (with some filtering, the amount of energy is low & the coupling high impedance) seems like a good idea (on first glance anyway).

[Bxaresearch]

If one is satisfied with fixed 1kHz and multiples, maybe an (series of) injection locked Wien bridges might be the way to go? There are some published designs with extremely low distortion at -140dB. For the synchronized injection locking, the harmonic content of a 1kHz pulse as mentioned above (with some filtering, the amount of energy is low & the coupling high impedance) seems like a good idea (on first glance anyway).

Thank you for this new idea I have not considered yet, I'll be looking at that

[Bxaresearch]

You're welcome.

This one bounced around my head a bit, so I've come for a second bite.  Setting aside the phase lock to incoming 1 kHz signal, there are some requirements to get coherent harmonics out...

• To operate PLL, require f_osc is a multiple of 1 kHz
• To get 2 kHz output, require f_osc is a multiple of 2 kHz
• To get 3 kHz output, require f_osc is a multiple of 3 kHz
• Thus to have the choice between (or multiple outputs) 2 and 3 kHz, f_osc must be a multiple of both 2 kHz and 3 kHz, i.e. a multiple of 6 kHz
  

As the number of harmonics of interest goes up, so too does the required f_osc.

Let's imagine we'd like the option to output the following: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 kHz
We can decompose those to products of primes: 1, 2, 3, 2², 5, 2*3, 7, 2³, 3², 2*5
Thus we could use f_osc = product(2³, 3², 5, 7) = 2520 kHz

Let's imagine we'd like to add on 11, 12 kHz
That's going to add some more products of primes: 11, 2²*3
Thus we could use f_osc = product(2³, 3², 5, 7, 11) = 27,720 kHz

And if we want to go a bit further and add on 13, 14, 15kHz
That's going to add some more products of primes: 13, 2*7, 3*5
Thus we could use f_osc = product(2³, 3², 5, 7, 11, 13) = 360,360 kHz

So if we want to use a single value of f_{osc[\sub] and output many different harmonics then fosc needs to be quite high.  Adjusting fosc ratio to vary depending on which harmonics we want to output could make a huge impact, though}

As the input clock frequency becomes impracticably high for generating the higher harmonics another approach might be to start out with a 1kHz train of narrow pulses. This will contain a high level of harmonics and each desired individual harmonic could be then picked out using a suitable narrow band-pass filter.

That one particularly interesting as the square wave output at 1kHz is already present and phase locket in DS360 and Sr1 (and equally in UPD and UPV analyzers). My task is basically get to the level of -130db distortion (minus noise) so that the distortion at very low output levels, even down to 1mV can be evaluated. This obviously requires different method than using an fft, but this part is not difficult as long as there is locked phase relationship between the fundamental and the square wave signal at harmonic frequency

[Bxaresearch]

In the locked wein bridge article quoted above the author says:
The oscillator powered up fine, giving a nice sinusoidal 5.15-kHz output at several volts, and independent EDN DI5291 measurements showed the second- and third-harmonic distortion products to be lower than −120 dBc


What measurement does it refer to? I failed to google it and don't know what that means

[Bxaresearch]

This is what was communicated to me by the article author:

You need calibrated notches of known frequency response.

My colleague did the actual distortion measurement.

He used RC notches.

 

To improve a source, you notch out its harmonics.

Then you notch out the fundamental after the DUT, so the spectrum analyzer doesn’t distort.

 

But in this case, we were measuring the source, so all you have to do is notch the fundamental, and know how much the notch loses at the 2nd and 3rd harmonics.

 

But these days, I think an SRS SR1 is good enough to get down below 120dB.

 

[jbb]

Sorry for delay... was ill over weekend.

So, I'm still trying to understand what the exact phase is required for.  Thinking about what's been in the thread so far, it seems like the plan is to:

• begin with excellent 1 kHz sine wave
• pass through some Device Under Test
• generate an n*1 kHz signal for use as a Local Oscillator (LO)
• mix the DUT output Vout with the LO
• low pass filter mixer output down to DC, i.e. select the nth harmonic
• measure said DC signal with multimeter to determine harmonic amplitude

Is that right?

If so, I can see why phase control is required; should the LO phase happen be 90 degrees away from the harmonic out of the DUT, you'll get zero.

With that in mind, how about this idea?

• begin with excellent 1 kHz sine wave
• pass through some Device Under Test
• generate an (n*1 kHz) - 55 Hz signal for use as a Local Oscillator (LO)
• mix the DUT output Vout with the LO
• band pass filter mixer output to select the 55 Hz output, i.e. select the nth harmonic
• measure said AC signal with multimeter to determine harmonic amplitude

Helpfully, you no longer need to fuss about DC offsets.  And you might be able to place the frequency of interest a bit higher than the 1/f noise corner of whatever post-mixer amplifiers you need.

I'm not certain that 55 Hz is exactly the right frequency offset here; I wanted to pick one that isn't an integer division of 1 kHz (to help mixer rejection) and lies between the 50 Hz and 60 Hz calibration points of the multimeter. You'd need to do some mixer harmonic analysis to check out all the mixer harmonic products.

I think in this case that you might be able to get away with an LO that isn't frequency locked to the 1 kHz tone, i.e. use 2 benchtop generators without clock synch.

[1audio]

This is an instrument used to "precisely" calibrate some of the lowest residual analyzers. http://www.shibasoku.com/download/avc/ac12b_e.pdf  When dealing with really low distortion all manner of little annoyances become challenges. The caps and resistors can limit the achievable distortion.
I would start with several of Victors oscillators, inexpensive and state of the art for distortion, typically -150 dB for all the harmonics. https://www.diyaudio.com/community/members/vicnic.19397/ 

The Shibasoku 725 uses a phase locked sampling system to measure harmonics and reject noise. Its pretty trick. And this software does something similar https ://www.data-odyssey.nl/Diana.html 

Also some power quality analyzers give level and phase of harmonics. They can be a big issue in high power transmission systems.  This may be of interest https://vitrek.com/pa910-high-accuracy-multi-channel-harmonic-power-analyzer/ Or at least you may be able to find out how they calibrate it.

[Bxaresearch]

Sorry for delay... was ill over weekend.

So, I'm still trying to understand what the exact phase is required for.  Thinking about what's been in the thread so far, it seems like the plan is to:

• begin with excellent 1 kHz sine wave
• pass through some Device Under Test
• generate an n*1 kHz signal for use as a Local Oscillator (LO)
• mix the DUT output Vout with the LO
• low pass filter mixer output down to DC, i.e. select the nth harmonic
• measure said DC signal with multimeter to determine harmonic amplitude

Is that right?

If so, I can see why phase control is required; should the LO phase happen be 90 degrees away from the harmonic out of the DUT, you'll get zero.

With that in mind, how about this idea?

• begin with excellent 1 kHz sine wave
• pass through some Device Under Test
• generate an (n*1 kHz) - 55 Hz signal for use as a Local Oscillator (LO)
• mix the DUT output Vout with the LO
• band pass filter mixer output to select the 55 Hz output, i.e. select the nth harmonic
• measure said AC signal with multimeter to determine harmonic amplitude

Helpfully, you no longer need to fuss about DC offsets.  And you might be able to place the frequency of interest a bit higher than the 1/f noise corner of whatever post-mixer amplifiers you need.

I'm not certain that 55 Hz is exactly the right frequency offset here; I wanted to pick one that isn't an integer division of 1 kHz (to help mixer rejection) and lies between the 50 Hz and 60 Hz calibration points of the multimeter. You'd need to do some mixer harmonic analysis to check out all the mixer harmonic products.

I think in this case that you might be able to get away with an LO that isn't frequency locked to the 1 kHz tone, i.e. use 2 benchtop generators without clock synch.

Sorry for the delay on my part. I was talking to some Analog Devices people (after reading the above mentioned article, beyond just the author himself) on this subject to see how the question of exceedingly low distortion and low amplitude is approached there. I am in no position to give an official run-down of what I found out, this is my distorted recollection, so take it as such.

What is special in the problem I'm considering is that the 1kHz sine wave is output at 1mV level and the distortion of the signal should be at least -130dB or preferably much lower. My current understanding is that there is no instrument that is able to generate and measure in this regime. Notice that all of the low distortion oscillators maintain the specified specification only at fairly high output level - 1V to 10V levels. One would think that attenuating this output is an easy task, however a resistive divider would quickly destroy the distortion specification of the oscillator. So what is done to address that is that the modestly clean sine wave generator, capable of outputting 2 channels that are phase locked is first attenuated and then harmonically filtered using proprietary circuits, until the wave is 1mV and less than -130dBc distortion. The other sine wave that is phase locked (but not exactly in phase at this point) is used to sync the lock-in amplifier. An additional step is to sent the said 1mV 1kHz signal through the DUT and then use another proprietary circuit to notch filter the fundamental (1kHz). The lock in amplifier can then be used to tune phase and frequency of each harmonic and accurately measure its amplitude.
That was my original plan, but I was lacking all of the proprietary aspects. So apparently my original question is answered by: no there is not a commercially made instrument that can accomplish all this out of the box. Pieces of the measurement chain must be built and care must be exercised during measurement and interpretation process.

[1audio]

130 dB below 1 mV? To avoid noise the impedances will be very low. I'm not sure how resistive dividers made with suitable quality resistors would add distortion. Metal film resistors added distortion is in the -160 dB range. You would need to notch the fundamental 40+ dB and then add an amp to boost the output of the notch to something a lock in amp can measure. You will also need to correct for the phase shifts from the notch filter. The DiAna program has a routine to do just that and will show the amplitude and phase of each harmonic. I suspect you will need an amp to get the levels high enough to process. This ADC and separate notch may be useful (and lots cheaper than a lock in amp) https://e1dashz.wixsite.com/index/cosmos

[tridac]

You ciould keep it simple and use audio transformer diode doublers, (full wave rectifier), or electronic split phase, with level detect / schmit trigger on the output. Low phase noise and guaranteed phase coherence with the input. Most likely less than $100 of parts. To get square wave, double frequency twice, then divide by two to get a symmetrical square wave output. So may ways to do that without costing the earth, or complexity...

[TimFox]

In the locked wein bridge article quoted above the author says:
The oscillator powered up fine, giving a nice sinusoidal 5.15-kHz output at several volts, and independent EDN DI5291 measurements showed the second- and third-harmonic distortion products to be lower than −120 dBc


What measurement does it refer to? I failed to google it and don't know what that means

That should mean that each of the second and third harmonic signals are more than 120 dB below the level of the fundamental signal.
This can be measured with a spectrum analyzer, easiest done if you first null out the fundamental with a distortion analyzer (e.g., -hp- 339A) to reduce the dynamic range required on the SA.
"dBc" means dB with respect to carrier.