EEVBlog #823 - Rigol DSG815 RF Signal Generator Teardown

[Ampere]

Great video, Dave. I was actually wondering how the signal was passed between those isolated modules right before you explained the hidden trace between the microvias. It's good to see something that's well-designed right after watching that tablet teardown.

Shame about those weird buttons, though. It seems like it was useless design step that didn't really add any value to the product. We've been using buttons that push straight down for a long time and nobody has ever said "I really wish that I could push my buttons to the side".

This is either the Batteriser fan club (1 member), the SJW white knight eunuch who has lost his mojo, or the german guy who made that fantastic (nearly) double sided PCB for the "medical tablet PC", the one with the Manhattan Beirut style construction.

Did I miss something? What does the SJW crowd have against Dave? Did he offend someone who identifies as a CapXon capacitor?

[c4757p]

Damn, a lot people don't like the buttons.

I'm curious about the output frontend. Looks like they have the attenuators before the output amplifier - is that normal? I'm no RF guy, but my intuition is that the SNR will be crap at low output amplitudes, with the signal almost swamped by amplifier noise. Shouldn't the attenuators come after the amplifier?

[retrolefty]

Looks like the video got pulled by Youtube.
Hope it's back soon.

Well That was BULLCRAP.
Why the Hell did they do that.
Oh Well, another company I will refuse to buy gear from be it used or new.
Go to Hell Rigol.

 Down girl, down. That some poster here speculated that Rigol had the video pulled is no reason for you to swear off Rigol just yet. What if they didn't do it all, which is more probable (in my speculation)?

 Facts can take time to appear but worth it in the long run. Hate can spread around the world before the truth can get it's pants on.

[G0HZU]

I had a closer look at some of the printed elliptic filters and I think the hardware may well support 3GHz operation?

Could the block diagram be closer to this one below?

If I assume the outer RF layer is 0.02" thick with a dielectric constant of about 4 then I think the first roofing lowpass filter after the bank of VCOs will cut off at about 3GHz and not 1.5GHz.

So could the 1.5GHz sig gen be crippled in firmware? Could the frequency plan for the block diagram be as below?
How thick are the outer layers of the PCB? Any images of the side view of the PCB?

All of the above is guesswork but the first ELP filter does look like a 3GHz design to me...

[EEVblog]

Well That was BULLCRAP.
Why the Hell did they do that.
Oh Well, another company I will refuse to buy gear from be it used or new.
Go to Hell Rigol.

Down girl, down. That some poster here speculated that Rigol had the video pulled is no reason for you to swear off Rigol just yet. What if they didn't do it all, which is more probable (in my speculation)?
 Facts can take time to appear but worth it in the long run. Hate can spread around the world before the truth can get it's pants on.

Yes, relax, it almost certainly wasn't Rigol.

[dr.diesel]

I'm curious about the output frontend.

Yeah, quick look only, but I suspect a similar market as the DSA815, mediocre performance, meh really.

[EEVblog]

Did I miss something? What does the SJW crowd have against Dave? Did he offend someone who identifies as a CapXon capacitor?

We had one SJW nutcase on this forum who has caused no end of problems and continues to stalk me and make my life hard (he's failed). Don't know if he's behind this, but he has threatened doing just this in the past, but it's the wrong flag if that's the case, this video was flagged as spam for some reason.

[c4757p]

Well, if G0HZU's diagram is accurate, the amplifier can be bypassed, which could give the best of two worlds: high amplitude output, and better linearity at lower amplitudes. If that's what they're doing, I totally missed that possibility, and totally approve.

[G0HZU]

I'm curious about the output frontend. Looks like they have the attenuators before the output amplifier - is that normal? I'm no RF guy, but my intuition is that the SNR will be crap at low output amplitudes, with the signal almost swamped by amplifier noise. Shouldn't the attenuators come after the amplifier?

I think the (GaN?) amplifier at the output is probably only there as a switchable booster amplifier when a high signal level is needed up at the top end of the sig gen's frequency range.

The loss in that suite of step attenuators up at a few GHz is probably enough to require a boost at high power levels at high frequencies?

It looks like there is another one of these GaN amps just before the suite of attenuator chips. But this one has the logo ground off it. Did someone forget to remove the logo from the final amplifier? Or maybe it is a replaced part (repair) on a demo unit and the repaired part hasn't had the logo removed?

Of all the parts on the signal path this output GaN amplifier is the one I would have expected to have the logo removed

It 'looks' like a Nitronex GaN part but the text/logo are not what I'm used to seeing. Not sure what it is unless it's some kind of lower spec clone of a Nitronex part?

[G0HZU]

Well, if G0HZU's diagram is accurate, the amplifier can be bypassed, which could give the best of two worlds: high amplitude output, and better linearity at lower amplitudes. If that's what they're doing, I totally missed that possibility, and totally approve.

Putting 'anything' after a decent ALC detector is a calibration risk on a wideband sig gen. In this case Rigol have fitted a string of attenuator chips (rather than a classic high performance mechanical attenuator with a flat frequency response) and also a switchable booster amplifier.

This introduces a lot of baggage wrt calibration so I suspect that this sig gen will have quite a complex calibration system. It will have to provide ALC correction factors across frequency that take into account the ripple/droop vs frequency in the attenuators and also the frequency response/ripple of the booster amplifier for when it is selected or deselected.

Also the quality of the AM can suffer because a classic ALC system on a sig gen will have feedback that can help to maintain a faithful AM modulation envelope even with a less than perfect amplifier. But that final GaN amplifier is outside the correction loop. So it will need to be quite well rated in terms of linearity if decent AM quality is desired

Also the output source impedance will probably not be that close to 50R across the range of that GaN amp. So the signal level (mismatch) uncertainty into less than ideal load impedances will probably be quite scary when the sig gen is in that mode with the booster amp selected.

[Monkeh]

BUt that 1000-2000h spec is for caps at 105^oC. Doesn't the lifetime get much longer at lower temperatures? I heard something like it doubles for each 10^oC lower.

Yeah, but for this equipment-price there are better suited (low imp & high reliability) caps on the market from Panasonic, Nichicon et al.

They are substantially more expensive.

Why must people get their panties in a twist over general purpose caps in applications where they are not stressed?

[rfeecs]

It 'looks' like a Nitronex GaN part but the text/logo are not what I'm used to seeing. Not sure what it is unless it's some kind of lower spec clone of a Nitronex part?

Wow.  I wouldn't have guessed they would use a GaN part for just 20dBm output power.  I suppose they are way backed off for decent harmonic performance.

It looks like it is this part based on the part number.  Seems like total overkill:
http://cdn.macom.com/datasheets/NDS-036r2%20NPTB00004A%20datasheet.pdf

[EEVblog]

Well, if G0HZU's diagram is accurate, the amplifier can be bypassed, which could give the best of two worlds: high amplitude output, and better linearity at lower amplitudes. If that's what they're doing, I totally missed that possibility, and totally approve.

Yes, that's likely what's happening. You wouldn't have a permanent amplifier after and attenuator in a performance instrument, that's dumb.

[G0HZU]

If that is the actual part they are using then they will be using it at a much lower drain voltage than 28V. I would question if it really is a Nitronex part. Could it be some kind of cheaper/derated clone with a similar part number?

[EEVblog]

Why must people get their panties in a twist over general purpose caps in applications where they are not stressed?

It's about best practice.
In a penny pinching low end scope, understandable. In a $2k-$3.6k instrument, not a good look.

[dr.diesel]

Why must people get their panties in a twist over general purpose caps in applications where they are not stressed?

I've replaced CapXon caps in stuff less than 5 years old, at "normal" temp conditions.

For $3600 (3Ghz model) using shit caps is inexcusable.   

[EEVblog]

Why is a 1000hr Nichicon more reliable than a 1000hr Capxon?

Better design and quality control.

[rfeecs]

If that is the actual part they are using then they will be using it at a much lower drain voltage than 28V. I would question if it really is a Nitronex part. Could it be some kind of cheaper/derated clone with a similar part number?

Doubt if it is a clone.
This thing is unmatched, with feedback only.  If they bias at 100mA, they would get a 5V peak swing into 50 ohms, giving 24dBm.  Minus losses.  So it's in the ball park after all.

[Monkeh]

Why must people get their panties in a twist over general purpose caps in applications where they are not stressed?

I've replaced CapXon caps in stuff less than 5 years old, at "normal" temp conditions.

For $3600 (3Ghz model) using shit caps is inexcusable.   

I wasn't talking about the CapXon caps.

[Alex Eisenhut]

Another decent quality piece of equipment marred by an offbeat, gimmicky, and in some ways childish, front panel design. This one is nowhere near as bad as the DS1054Z or the ridiculous DP832, but I still think they need a change of personnel in that department.

They're stylistically meant to be that way. So that it rhymes.

[crispy_tofu]

Damn, a lot people don't like the buttons.

I don't know about you, but those buttons on the left of the unit really irk me. Why are they italicised? I think they went a bit too far on the design... 

[halexa]

Hi,

I notice this RF-Voodoo thingy in the video, firts i thought it was the Open source hardware loggo. Just wondering if any one knows what it does?

[f4eru]

I notice this RF-Voodoo thingy in the video, firts i thought it was the Open source hardware loggo. Just wondering if any one knows what it does?

That's a coaxial test pad.
see page 9: http://www.ingun.de/media/pdf/Werbemittel/INGUN_RF_Catalogue_EN.pdf

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...

[G0HZU]

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.

[G0HZU]

I had a look through this Rigol datasheet and there are a few clues wrt the frequency plan there.

http://beyondmeasure.rigoltech.com/acton/attachment/1579/f-0636/1/-/-/-/-/DSG800_DataSheet_EN.pdf

eg it lists 5 bands on the second page and this gives a clue as to the division ratio associated with each range.

So I tried to amend my block diagram to include this info. See below for another attempt to produce a frequency plan for the 3GHz version. I still think the same hardware may be used for the 1.5GHz version. However, one thing I can't fully explain yet is the phase noise plot they provide on page 3.

The traces for 3GHz and 1GHz make sense because there is a divide by two on the 1GHz range plus it will use a VCO running at 2GHz. So you can expect the phase noise to be 6dB lower because of the divide by two and maybe another 4 dB lower because of the 2GHz VCO. So the phase noise at 1GHz is about 10dB better than the phase noise at 3GHz.

But I can't explain the phase noise plot at 100MHz.It seems lower than I expected and this implies there are some more stages of division somewhere. Otherwise, I'd expect the phase noise profile at 100MHz to only be a couple of dB lower than the phase noise profile at 1GHz if my block diagram is correct.

The table on page 2 only quotes a divide by 4 for 100MHz so this agrees with my block doagram but this means a cleanup of (only) about 12dB wrt the phase noise at 3GHz and then the BFO is used to downconvert to 100MHz.

So something doesn't add up here because the cleanup at 100MHz is more like 24dB. I'll try and spend some time looking at the other side of the RF board to try and spot an extra divider (divide by 16?) and an associated lowpass filter. The phase noise plot suggests that 100MHz is achieved with a simple divide by 16 rather than the  'divide by 4 plus BFO' shown in my block diagram.