Not all testsets have samplers, some have mixers. If it is a 8517 it has mixers, not samplers. But, in any case there are 4 freuqency converters in the 8510.
" Not all testsets have samplers, some have mixers. If it is a 8517 it has mixers, not samplers. But, in any case there are 4 freuqency converters in the 8510. "
I believe that should read 85110 (pulsed test set). The mm-wave test sets also have mixers. From what I recall, all the other test sets have samplers. T/R test sets have three converters and S-parameter test sets have four.
What, exactly, is the difference between a 'mixer' and a 'sampler'? They both convert one RF signal to baseband by multiplication with another, don't they?
Is a sampler just a general term for a diode mixer with an IF range extending down to DC?
" What, exactly, is the difference between a 'mixer' and a 'sampler'? They both convert one RF signal to baseband by multiplication with another, don't they?
Is a sampler just a general term for a diode mixer with an IF range extending down to DC? "
In our vernacular, sampler is used to mean a high-harmonic mixer, generally with a very short conduction angle. For example, the conduction angle on the 8510 sampler is set by a shorted line across the LO drive port. Samplers are typically driven with pulses, not sine waves, and the get their high frequency response from the rise time of the pulse (NOT, as is so often mistakenly quoted, from the pulse width).
Mixers are driven at the LO port with a sine way, single balance mixers have a conduction angle nearly equal to 180 degrees, double balanced mixers are nearly 360 degrees.
With high conduction angles come higher efficiencies, or lower noise floor. A subtlety is that a sampler can have a high voltage conversion efficiency, but at the cost of having a very high output impedance, meaning it will have an equally high kTB noise. For a sampler, the noise figure degrades as the log of the harmonic content of the pulse. You can almost think of it as each harmonic converts its surrounding noise spectrum into the IF, so if you have 100 harmonics, you have 100 times the noise, or 20 dB more.
Thanks, Dr_joel -- I appreciate the info. I know the question is somewhat OT for the forum. I have been playing around with an "eBay special" 8753A for a few days, trying to understand a little more about its front-end architecture than the manual describes. The samplers have at least a half-dozen connections to the PC board, making me wonder if there was a lot more inside them than just a passive diode mixer.
I will confess that I don't fully grok the effect of conduction angle as it applies to a pulse-driven mixer. As you suggest, the edges are where the mixing process actually happens; I'm guessing the effects of shorting out the "slow parts" of the waveform are to improve bandwidth by keeping the diodes out of saturation, and/or to reflect any stray IF energy back into the diodes? The former doesn't sound like an effect the 8753 samplers could take advantage of, since they have to work with all of the LO harmonics plus the fundamental at 30-60 MHz.
I've noticed experimentally that feeding a comb to an ordinary Mini-Circuits DBM results in impairment of the conversion-loss spec at each picket roughly equal to that harmonic's amplitude relative to the LO drive spec. So if I were to use a DBM as a sampler, the effective NF at harmonic N would be the conversion loss at that harmonic plus the same 10*log(N) effect you're describing, right? Isn't that just as true of a sampler? Ultimately, I'd like to understand the pros and cons of using an inexpensive DBM module as a sampler in a PLL, and to get there, I need a better understanding of what the differences actually are.
What you're saying also makes me wonder if the 8753's samplers have a ~3-GHz LPF on either the LO or RF ports or both. The SRD generator emits strong harmonics well beyond 3 GHz, so I imagine you don't want to pay the 10*log(N) NF penalty on harmonics that will never be used.
Hello John, it's actually quite a bit more complicated with respect to noise figure than my first post indicated.
In the 8753, the sampler includes first in common base input buffer amplifier, where there is a series 44 ohm resister going to the emmiter of a Si transistor. This provides reverse isolation of the sampler to prevent other products from spewing back out the input port, as well as providing a good input return loss. The collector of this transistor goes through a meander line to a diode pair (hooked up much like a single balanced mixer). The pulse is fed through a balun hung across what would be the LO port of the SBM. The IF port goes to a single stage amplifer for the IF. In addition, there are some DC inputs that let the mixer diodes be turned fully on, which is how the low band (16 MHz and below) signal gets to the second converter. The second converter converts either 1 MHz (sampler IF) or 30kHz-16 MHz (low-band IF) to 4 kHz. The fundamental VTO is 30-60 MHz, but we divide it by 2 for 15-30 MHz, before it goes into the pulser. This 4:1 ratio of pulse repetition can cause some conversion efficiency droop at low frequency. In addition, there is self-biasing diodes and capcitors on the LO feed that help turn on the diode and provide a more constant RF-IF conversion with varying LO drive levels (LO varies due to temperation, and pulse-gen frequency).
As for noise figure, the most proper (in my opinion) way to look at the sampler as providing a high Z input into the IF. The value of the Z is the time averaged impedance of the turn on for the sampler. Wider conduction angle (longer turn on times) will reduce the noise, but also limit the bandwidth, due to a spectral nulling effect of the falling edge.
There is no LP on either the RF or LO inputs, other than the natural tendency to have the buffer amp and balun roll-off.
I spent about 2 years of my life working on that sampler, taking over after 2 other engineers, so there are an awful lot of little strange things that ended up in the design. A couple of key aspects, that we don't care about any more, is that it was to have a raw conversion efficiency of +- 1 dB to 1.3 GHz, and an input match better than 26 dB. Now, we just error correct everything, but at the start of the project, it wasn't clear that we would even have error correction built in.
Thanks for the insights into those mysterious gold boxes, Joel, and for the explanation of the noise model. I've got a better picture of it in my head now.
The isolation approach is definitely interesting. I'd previously opened the comb generator module for a look and noticed that there was only one SRD driving the three output jacks with a simple resistive(?) splitter. So I couldn't imagine how you were getting > 100 dB of input isolation between samplers with the LO ports effectively in parallel with each other. Seemed pretty clear that there would need to be one, or probably more, active isolation stages in the samplers.
Before I opened the CG module, I was pretty sure that each sampler would have its own SRD, just because it'd be easier to isolate the low-frequency SRD drive signals from each other. You guys definitely didn't take the easy way out on this one.
I'd also forgotten about the need to turn the diodes on for the LF/HF signal path where the SRD isn't used, but yes, that makes sense and would account for at least one or two more pins. (I'm a little unclear as to how 15-30 MHz pulses yield comb harmonics at 30-60 MHz intervals, though.) I just ordered a few SRDs from ASI and look forward to playing with them on the bench when I have some time...
I believe that should read 85110 (pulsed test set). The mm-wave test sets also have mixers. From what I recall, all the other test sets have samplers. T/R test sets have three converters and S-parameter test sets have four.
Is a sampler just a general term for a diode mixer with an IF range extending down to DC?
In our vernacular, sampler is used to mean a high-harmonic mixer, generally with a very short conduction angle. For example, the conduction angle on the 8510 sampler is set by a shorted line across the LO drive port. Samplers are typically driven with pulses, not sine waves, and the get their high frequency response from the rise time of the pulse (NOT, as is so often mistakenly quoted, from the pulse width).
Mixers are driven at the LO port with a sine way, single balance mixers have a conduction angle nearly equal to 180 degrees, double balanced mixers are nearly 360 degrees.
With high conduction angles come higher efficiencies, or lower noise floor. A subtlety is that a sampler can have a high voltage conversion efficiency, but at the cost of having a very high output impedance, meaning it will have an equally high kTB noise. For a sampler, the noise figure degrades as the log of the harmonic content of the pulse. You can almost think of it as each harmonic converts its surrounding noise spectrum into the IF, so if you have 100 harmonics, you have 100 times the noise, or 20 dB more.
I will confess that I don't fully grok the effect of conduction angle as it applies to a pulse-driven mixer. As you suggest, the edges are where the mixing process actually happens; I'm guessing the effects of shorting out the "slow parts" of the waveform are to improve bandwidth by keeping the diodes out of saturation, and/or to reflect any stray IF energy back into the diodes? The former doesn't sound like an effect the 8753 samplers could take advantage of, since they have to work with all of the LO harmonics plus the fundamental at 30-60 MHz.
I've noticed experimentally that feeding a comb to an ordinary Mini-Circuits DBM results in impairment of the conversion-loss spec at each picket roughly equal to that harmonic's amplitude relative to the LO drive spec. So if I were to use a DBM as a sampler, the effective NF at harmonic N would be the conversion loss at that harmonic plus the same 10*log(N) effect you're describing, right? Isn't that just as true of a sampler? Ultimately, I'd like to understand the pros and cons of using an inexpensive DBM module as a sampler in a PLL, and to get there, I need a better understanding of what the differences actually are.
What you're saying also makes me wonder if the 8753's samplers have a ~3-GHz LPF on either the LO or RF ports or both. The SRD generator emits strong harmonics well beyond 3 GHz, so I imagine you don't want to pay the 10*log(N) NF penalty on harmonics that will never be used.
it's actually quite a bit more complicated with respect to noise figure than my first post indicated.
In the 8753, the sampler includes first in common base input buffer amplifier, where there is a series 44 ohm resister going to the emmiter of a Si transistor. This provides reverse isolation of the sampler to prevent other products from spewing back out the input port, as well as providing a good input return loss. The collector of this transistor goes through a meander line to a diode pair (hooked up much like a single balanced mixer). The pulse is fed through a balun hung across what would be the LO port of the SBM. The IF port goes to a single stage amplifer for the IF. In addition, there are some DC inputs that let the mixer diodes be turned fully on, which is how the low band (16 MHz and below) signal gets to the second converter. The second converter converts either 1 MHz (sampler IF) or 30kHz-16 MHz (low-band IF) to 4 kHz. The fundamental VTO is 30-60 MHz, but we divide it by 2 for 15-30 MHz, before it goes into the pulser. This 4:1 ratio of pulse repetition can cause some conversion efficiency droop at low frequency. In addition, there is self-biasing diodes and capcitors on the LO feed that help turn on the diode and provide a more constant RF-IF conversion with varying LO drive levels (LO varies due to temperation, and pulse-gen frequency).
As for noise figure, the most proper (in my opinion) way to look at the sampler as providing a high Z input into the IF. The value of the Z is the time averaged impedance of the turn on for the sampler. Wider conduction angle (longer turn on times) will reduce the noise, but also limit the bandwidth, due to a spectral nulling effect of the falling edge.
There is no LP on either the RF or LO inputs, other than the natural tendency to have the buffer amp and balun roll-off.
I spent about 2 years of my life working on that sampler, taking over after 2 other engineers, so there are an awful lot of little strange things that ended up in the design. A couple of key aspects, that we don't care about any more, is that it was to have a raw conversion efficiency of +- 1 dB to 1.3 GHz, and an input match better than 26 dB. Now, we just error correct everything, but at the start of the project, it wasn't clear that we would even have error correction built in.
The isolation approach is definitely interesting. I'd previously opened the comb generator module for a look and noticed that there was only one SRD driving the three output jacks with a simple resistive(?) splitter. So I couldn't imagine how you were getting > 100 dB of input isolation between samplers with the LO ports effectively in parallel with each other. Seemed pretty clear that there would need to be one, or probably more, active isolation stages in the samplers.
Before I opened the CG module, I was pretty sure that each sampler would have its own SRD, just because it'd be easier to isolate the low-frequency SRD drive signals from each other. You guys definitely didn't take the easy way out on this one.
I'd also forgotten about the need to turn the diodes on for the LF/HF signal path where the SRD isn't used, but yes, that makes sense and would account for at least one or two more pins. (I'm a little unclear as to how 15-30 MHz pulses yield comb harmonics at 30-60 MHz intervals, though.) I just ordered a few SRDs from ASI and look forward to playing with them on the bench when I have some time...