EEVBlog #823 - Rigol DSG815 RF Signal Generator Teardown

[f4eru]

I wonder what U801 is (the temperature stabilized part marked TKK)
Perhaps it's a RSSI meter so the power can be regulated to a precise value...

My guess is that the 6 pin package just contains a couple of matched detector diodes.

One will be used as the RF envelope detector diode for the ALC and the other will be used as part of the detection compensation. i.e. it will be used to cancel out the Vdrop in the RF diode over temperature. The oven will presumably be used to make this diode compensation scheme even more stable.

Could be, but there are also nicely integrated log amplifiers/detectors, with 70dB dynamic range.
for example : http://www.analog.com/en/products/rf-microwave/rf-power-detectors.html#envelope-peak-detectors
Also, it's referenced "U" for IC, not "D" as a diode pack, but that's just supposition.

What's the dynamic range to be expected from a nicely compensated diode detector ?

[G0HZU]

It's not going to be a logamp or one of the true rms detectors because it will almost certainly need to be used for the AM modulation as well as the ALC. So it will need to be able to track the modulation envelope quite accurately and produce an accurate demodulation waveform to feed back to the error amplifier in the ALC system. So it needs to be very, very fast and it needs to track the outer envelope of the AM waveform very accurately

In AM mode the wanted baseband AM modulation waveform is typically fed to the reference pin of an error amplifier in the ALC feedback loop. So the ALC detector diode will try and track the RF envelope and ensure its detector output is fast enough to track the baseband waveform. This is exactly what you want for generating good quality AM modulation

I think it will be a classic dual diode from someone like Agilent/Avago. eg one where the two diodes are matched but not connected inside the 6 pin package. eg something like the classic HSMS-28xx series?

 

[G0HZU]

What's the dynamic range to be expected from a nicely compensated diode detector ?

I guess it depends on the complexity of the design but the classic way to provide AM and ALC in a lab sig gen is to use a decent diode based envelope detector. This method dates back many decades and a classic example is the HP8640 lab sig gen from the early 1970s. I think that all of my various sig gens I have here use this method. I expect Rigol will adopt a similar system.

[G0HZU]

I had a rummage through a few diode detector datasheets and I think the ALC detector (marked as TKK) could be a HSMS-286K dual diode from Agilent/Avago?

[G0HZU]

I also had a closer look at the big square chip (DDS?) and this looks like it may well be a dual DAC rather than a DDS.

I also missed a divide by two prescaler at its clock input so could it be an AD9747 Dual 16 bit DAC from Analog Devices? The pinouts look similar at a glance and it appears to have dual outputs so maybe one output feeds the main PLL as an agile reference signal and the other gets used for LF signals?

I wonder if Rigol are overclocking it a bit because it looks like it is getting clocked at something like 2600/8 = 325MHz.

It's providing the same functionality as a DDS in terms of producing an agile (small stepping) reference for a hybrid PLL but it looks like the Spartan chip can control this DAC fast enough to provide an HF reference signal with fast modulation rates etc. Impressive stuff indeed...

[rfeecs]

The video clearly shows the VCO is a CRO3640B.  Dave even reads out the part number.  Not sure why he pulled up the wrong data sheet, except that there is none on the web site for the right part number.  There is a CRO3640A data sheet.  The frequency is 3640MHz:
https://www.zcomm.com/pdfs/datasheets/CRO3640A-LF.pdf

[f4eru]

I guess it depends on the complexity of the design but the classic way to provide AM and ALC in a lab sig gen is to use a decent diode based envelope detector.

I didn't know that the proper way to generate quality AM is to implement an AF closed loop through a detector. Nice to know

[G0HZU]

The video clearly shows the VCO is a CRO3640B.  Dave even reads out the part number.  Not sure why he pulled up the wrong data sheet, except that there is none on the web site for the right part number.  There is a CRO3640A data sheet.  The frequency is 3640MHz:
https://www.zcomm.com/pdfs/datasheets/CRO3640A-LF.pdf

Good spot...

I lazily went with Dave's datasheet that showed ~ 2600MHz. This solves one issue I had with the 3GHz variant. It didn't seem to be a good idea to have a permanent VCO that sat inside the 3GHz tuning range of the sig gen because this can lead to beating effects if the sig gen is tuned to a nearby frequency.

So with this VCO up at about 3640MHz I also had to revisit what was fitted as the big mystery DAC. This is because I now think it will be clocked at 3640/8 = 455MHz.

Maybe it is something like an Analog Devices AD9122 DAC?

[GK]

At a rough guess how close would the BOM for all PCB parts including all specialist RF ICs bought in one off/single-figure-small quantities compare to RIGIOLs retail price for the complete instrument?

[AF6LJ]

The video clearly shows the VCO is a CRO3640B.  Dave even reads out the part number.  Not sure why he pulled up the wrong data sheet, except that there is none on the web site for the right part number.  There is a CRO3640A data sheet.  The frequency is 3640MHz:
https://www.zcomm.com/pdfs/datasheets/CRO3640A-LF.pdf

Good spot...

I lazily went with Dave's datasheet that showed ~ 2600MHz. This solves one issue I had with the 3GHz variant. It didn't seem to be a good idea to have a permanent VCO that sat inside the 3GHz tuning range of the sig gen because this can lead to beating effects if the sig gen is tuned to a nearby frequency.

So with this VCO up at about 3640MHz I also had to revisit what was fitted as the big mystery DAC. This is because I now think it will be clocked at 3640/8 = 455MHz.

Maybe it is something like an Analog Devices AD9122 DAC?

Nice job on reverse engineering this.

[ktulu]

Did anyone noticed the 3 LED looking diodes in the VCO section? (D200, D205, D210)

[G0HZU]

Nice job on reverse engineering this.

Thanks   I can only guess at all of this though because I don't have access to the hardware. So there are bound to be mistakes.

Here's an updated block diagram although this block diagram still doesn't explain the very low phase noise on the 100MHz trace on Rigol's datasheet. So something still isn't right somewhere.

But the 5 bands shown in my block diagram agree with the 5 frequency bands shown in their datasheet in terms of frequency range and divsion ratio.

Did anyone noticed the 3 LED looking diodes in the VCO section? (D200, D205, D210)

Presumably these wink on and off as each VCO gets selected. This would help an engineer during the dev phase.

The transistor used for the VCO appears to be an Infineon part = BFP450. I have a non linear model for this part here and I simulated the VCO layout for each range using this part. I adjusted the C-E capacitor for best negative resistance and the little printed resonator and the two series varactor diodes do seem to deliver the required oscillation frequency to cover about 1800MHz to 3600MHz in three ranges if I use typical capacitance values for this type of VCO tuning diode.

I'm estimating that the 50 ohm traces are about 0.04" wide and the outer RF layer is somewhere around 0.02" thick with a dielectric constant around 4 to 4.3.

The elliptic lowpass filter marked as 3000MHz in my block diagram does seem to cut off at just over 3GHz if I simulate it in Sonnet using the PCB data above.

This (and the VCO resonator dimensions) does suggest that this hardware is designed to cover up to a 3GHz frequency range.

So maybe the 1.5GHz version is just crippled in firmware? So maybe it can be hacked to become the 3GHz version?

[G0HZU]

If Dave still has the sig gen I can suggest some (spurious) tests to see if the block diagram is valid? i.e. places to look for unwanted spurious effects of this frequency plan.
The sig gen doesn't have to be taken apart again to do these tests. Anyone can do them if they have the generator and a spectrum analyser.

[rfeecs]

The transistor used for the VCO appears to be an Infineon part = BFP450. I have a non linear model for this part here and I simulated the VCO layout for each range using this part. I adjusted the C-E capacitor for best negative resistance and the little printed resonator and the two series varactor diodes do seem to deliver the required oscillation frequency to cover about 1800MHz to 3600MHz in three ranges if I use typical capacitance values for this type of VCO tuning diode.

Awesome reverse engineering.  Maybe you're becoming a little obsessed?
I kind of wonder why they bought the Z-Comm VCO instead of making their own.

I'm estimating that the 50 ohm traces are about 0.04" wide and the outer RF layer is somewhere around 0.02" thick with a dielectric constant around 4 to 4.3.

The elliptic lowpass filter marked as 3000MHz in my block diagram does seem to cut off at just over 3GHz if I simulate it in Sonnet using the PCB data above.

This (and the VCO resonator dimensions) does suggest that this hardware is designed to cover up to a 3GHz frequency range.

So maybe the 1.5GHz version is just crippled in firmware? So maybe it can be hacked to become the 3GHz version?

It does look like it.  It seems bizarre that they sell the same unit for 1.8 times the price for 3GHz vs 1.5GHz.  But they do that with oscilloscopes, too.  I find it borderline morally bankrupt.

[G0HZU]

Maybe you're becoming a little obsessed?
I kind of wonder why they bought the Z-Comm VCO instead of making their own.

I've designed a few xGHz VCOs in my time but nothing as good as that Zcomm unit in terms of phase noise. It does produce very low phase noise
It needs to have extremely low phase noise because it is required to provide the low noise clock for the DAC and also the BFO LO. There will be 12dB cleanup in the divide by 4 for the 910MHz BFO and this will mean that the phase noise of the BFO will be insignificant wrt the phase noise from the discrete (~3GHz?) Rigol VCO #3 after it has been divided by 4.

If I chuck some typical design parameters into a basic oscillator design equation then this shows that this is a special design by Zcomm. It helps that it is a narrowband VCO but I suspect that I would struggle to design a VCO that could get within 10dB of that phase noise response. I could easily manage to get within 15dB and I'd expect to get within about 10dB but I've never designed anything that clean up at 3.5GHz before.

It's tempting to try but I think I'd end up using more DC input power and more space even if I got close to the Zcomm spec...

[KE5FX]

Yep, rolling your own VCOs is not a good plan in any commercial design unless your name is Rohde or Leeson or Scherer or something like that.  Companies like ZComm and Synergy Microwave have put decades' worth of R&D into those things. 

Looks like the signal path for 100 MHz output is completely different; they aren't just dividing the output from a UHF synthesizer block.  There's a PLL shoulder at 10 kHz that's completely absent in the 100 MHz trace.

Also, I would expect them to use an offset mixing scheme in their hybrid PLL instead of multiplying a DDS... but then, I'd expect lower PN if they did that.

[G0HZU]

Yes, I can't explain the odd looking phase noise trace at 100MHz.

Maybe there is an extra range squeezed in that covers a segment around 100MHz. Maybe this feeds direct from the DAC and bypasses the system that uses the 3 VCOs and the dividers. So the phase noise would be better in this sweetspot band of frequencies.

The signal path that has the AM/ALC attenuators looks like it has blocking caps designed for VHF and above so I'd expect the BFO circuit will be used for 9kHz through to several tens of MHz at least. But maybe the next range up will be this DAC/DDS driven range.

At work, we stopped designing our own VCOs (that go into saleable products) many years ago. The cost is way too high because of all the NRE time/costs in order to prove the design over temperature and shock and supply/load pull etc etc.  However, you do still need to know a bit about VCOs when selecting a suitable VCO because even the commercial ones will sometimes have 'gotchas' built into the design. With experience it is possible to look through a VCO datasheet and see what is 'missing' or different in order to predict what these gotchas are.

I do enjoy designing oscillators up in the microwave bands although these days I only design them as test oscillators.