RFPA noise blanking...

[aot]

Hi,

I am currently building a high power (ideally 100W+ CW) noise blanking gate for the output of an AMT M3205A power amplifier for use in MRI. Basically, even with the blanking circuitry switched in (for all stages in the amplifier) there is a significant amount of noise being emitted from this amplifier, and this is easily seen in the MRI images. The frequency of operation for this is 123.259MHz.

I've got a basic design which is essentially a series/shunt type SPDT switch using PIN diodes, except with parallel resonance circuits around each diode to increase isolation (see schematic at the bottom - this is the basic circuit and the component values change quite a bit, I think the ones in the schematic are what I found simulated well). In QUCS I have built a simulation which works well, but I just can't get good performance for the real thing, especially when driving it with a bit of power. Above about 10W it is reflecting so much back the amplifier goes into overdrive and shuts down!

I built something similar to this for blanking some 1W amplifiers, and got very good results. However the Tx path was in series with a PIN diode forward biased, in this high power version the Tx path is via an inductor in parallel to the reverse biased PIN. It's built as a microstrip layout on FR4, but I've had to jiggle things around and cut the tracks as the board was originally for the 1W version (I don't have access to PCB milling machines here), so there's probably some parasitics there causing a mismatch.

I keep modifying it, but basically the problem I'm having is that the match of the Tx path (via D2) is poor, and this mismatch gets worse with more power from the network analyser.

The former is because of parasitics from the board, D2, C3 etc which means that the Tx path is not purely through C1, C2, L1 and C4, so consequently C4 cannot cancel L1's reactance. I think by working methodically through it all I can figure out the sources, but I'm not sure how to cancel it out. I did build a L matching network on the RF out port (P6) and managed to get the match to 50ohms, but is this the right thing to do?

As for the worse performance with higher powers this could be a few things. My NA might not be calibrated (it's about 15 years old and hasn't seen a service for years), but I recently compared it with some others and found it to perform adequately. There could be some non-linearities in the passive components I am using, but I think this is unlikely as they are decent P90 capacitors from passive-plus (1111P series), and the inductors are hand-wound air-gap, from 1.25mm diameter enamelled copper. The RF chokes aren't fantastic, and have since I've replaced them with parallel resonant traps and get 12k+ from them, which I would have thought would be sufficient.
I think the issue is that the diodes are only reverse biased to 5V, so there's a possibility the Tx power is turning them on, but this would be quite a change that I would need to find a higher voltage dual rail power supply! The PIN's I am using are passive plus PPD1200's (most of my components are non-magnetic types for MRI, this isn't necessary here as the gate isn't going inside the MRI scanner, but I happen to have all the bits already).


I'm a Ph.D student, and my background is Physics, so everything RF is what I have learnt over the past four years, and for the most part that has been building MRI coils and associated low power bits. Also the focus of my project is using that noisy amplifier in the imaging, so I really want to just be able to get this done and move on, but I appreciate that this is probably not a trivial circuit, especially given my limited knowledge, expertise and equipment, and that high power RF is particularly tricky!

Any advice would be particularly useful, I've got some more structured questions below, but general comments would be really appreciated.

1. The matching network I mentioned, is that the right way to sort out the parasitic reactances and get a better match, or should I be series resonating/shunting them out first and then build a matching network if necessary?

2. is this power non-linearity I'm seeing due to the diodes not being reverse biased enough, or is it just due to the design and I should try to account for it when building the matching network.

3. Is this even a good way to build a RF switch, I've had a look on the net and seen some TR switches, many are the series/shunt type T/R switches, perhaps I should be going with those - power handling was my concern though.


Many thanks,

Aaron

Ωµ

 

[kspalla]

can you re-share the picture/schematic?
i understand its better to use control on final PA to improve overall efficiency, but its complex. simpler way is to control driver or pre-driver, have you done study on these methods?

 

[aot]

the schematic thumbnail appears to be broken, but click on the dark space.abed out will show a full size image.

I don't follow what you mean by control. I'm not trying to build a new rfpa, I already have an extremely linear one specifically designed for nmr/mri, the only issue is noise. The rfpa has blanking on all stages, which is controlled by a ttl level pulse, however even with this in place the noise at the output is still seen in an MRI image (a different coil is used for reception and so noise injected into the system is picked up by it).

 

[kspalla]

control means blanking in this case.
why so many 50R? i guess some thing to do on sch improvement.

 

[biff44]

why is c4 only 19pf? Also, make you +5v supply something like +20V. Who's pin diodes are you using.

 

[kspalla]

in my view +5 to +20V change may not be effective but -5 to -10 or so. this is because of the diode direction.

 

[tony_lth]

Hittite Microwave
**broken link removed**
I think TR switch is better than PIN diode SPDT. You can control driver, and you can control the final PA power supply.

 

[aot]

The 50R after the chokes is to provide current regulation to bias the diodes at 100mA. The other 50R is a load for in black mode.

Pin diodes are from Passive Plus, Inc. - Passive Plus Inc., a manufacturer of HI-Q Capacitors and other RF/Microwave and MRI components.. Probably not ideal but I don't have a lot of choice in components because I have no funding left.

Yes in theory C4 should be 22p, but I found in simulation with parasitics the actual value to be 19pF, and on the board I am using 20pF as this is the value that seemed to be most effective.

I'm thinking of changing the reverse bias voltage to +15 v (I have easy access to +-5 and +-15), to go much higher would require me building a higher voltage supply, which will take time, something I'd rather not spend.

I have looked at the minicircuits offerings in the past, but I need something capable of 100W CW, I couldn't find anything high power. also I have no money to
spend!

In MRI I need to very quickly switch between blanked and transmit, on the order of every 20ms, i.e. one pulse lasting 500us every 20ms, so I don't think I can turn on and off the PA quickly enough and retain linearity.


You mention T/R switches, any specific topology?

 

[biff44]

No, what I mean is that C4 is a DC blocking capacitor. It should pass the rf signal un-attenuated, but block the DC bias. It should have an reactance of perhaps -j0.1 ohms or so.

A 19 pf cap at 123 MHz has a reactance of -j68 ohms. So 70% of your input power is reflected by this capacitor alone! Make it 10,000 pf. While you are at it, make C2 bigger too.

Make sure ALL of your capacitors have a big enough breakdown voltage to handle 100 watts!

Your bias network inductor H1 is theoreticaly big enough (1 uh), but does it really have 1 uh inducatance at 123 MHz? If it is a typical 1 uh inductor, it will resonate at a lower frequency than 123 MHz. If so, it might not be blocking your RF signal anymore.

 

[aot]

It's not just a DC blocking cap, it also series resonates out the reactance of L1. Theoretically, the combined reactance of L1 and C4 at 123.259MHz should be zero (and in simulation with ideal components it is), of course due to parasitics it doesn't quite get there. The problem I am having is that there are other parasitics in the circuit that I need to get rid of to improve the match at the output.

 

[biff44]

I do not have time to analyze this for you, but I think you are not using correct values for componenets in your RF analysis. As I said, parasitics ARE important. If you put 1 uh in your program to analyze and inductor, but it is instead acting like a capacitor because you are above resonant frequency, then you analysis is going to deviate from real world test results! Due to the nature of the application, these components are going to be phyically large, and therefore will have significant parasitic effects. Start by actually measuring the components to see what they are doing.

I assume you are using the PPD1200 pin diode? If you look at the data sheet, you can tell that this is one mondo big PIN diode. It has a reverse biased junction capacitance of 3.2 pf at 50 VOLTS! You are only reverse biasing it to 5 volts. Do you know the junction capacitance at only 5 v reverse bias? If you assume it is still only 3.2 pf....then you probably still believe in fairy tales. You really need to bump up that power supply voltage OR measure the capacitance at 5 V and use that in your analysis (although I predict you will not like what you see at 5V!)

If it were me, I would make C3 AND C4 really big, and only use L1 to anti-resonate the D2 when it was reverse biased. In forward bias, D2 only has .3 ohms, so it will short out L1 and not need any other reactances. And I would use at least 25 volts for both + and - supplies (preferrably 50v).

Then do the same to the other diode's circuit.

 

[aot]

Thanks for your comments bif44

In my analysis I have used parasitic values for my capacitors given on by the ATC100B datasheets (equivalent capacitors to what I am using), I have measured the SRF of my inductors and found them to be all above 500MHz. My chokes have a SRF of about 220MHz, however I have now moved on to LC trap circuits to block the RF because I can get about twice the blocking impedance (13k+). I am however not relying on the results of the simulation in selecting component values and setting up the real-life circuit, my point about the simulation is that theoretically it can provide good isolation etc.

My main concern with doing as you suggested is power handling, otherwise I would go with that method because I've built two that operate in that way and work well handling only 1W. Would D2 be capable of handling over 100W CW going through it, I'm not so sure it would.

 

[biff44]

I am not sure of your question. Are you in fact passing 100w RF through the pin diode D2 when it is forward biased? Or are you doing things backwards and somehow trying to pass power thru it when d2 is reverse biased by some sort of resonance with the inductor?

I will assume you are trying to use the switch normally, and D2 is forward biased when you are passing RF. The data sheet shows 0.3 ohms rf series resistance with 10 ma dc bias. This is pretty good for power handling. Let me explain. You are giving the diode more like 80 ma dc bias, so the Rs will get even lower (lets assume it gets to 0.15 ohms). The only thing that would keep that D2 from passing 100 watts would be if it melted down (or maybe arced over). To melt down, it would probably have to reach above 125 deg C. So, how hot is it?

P=V*I =(I*Z)*I
Rearranging: I =√(P/Z) = √(100 watts/50 ohms) = 1.14 Amps rms

The diode series resistance is assumed to be 0.15 ohms, and the current is above. So the actual power dissipated in the diode is

Pdiss=Rs*(Irf)²=(0.15 ohms) * (1.14 amps)²=0.3 watts of heat

The diode has a 15 deg C/watt thermal resistance, so the diode temperature is:
T = 15 deg C/W * 0.3 W = 4.5 deg C.

You will have some temperature rise thru the board material too, since the diode is series mounted on top of the board, but it is pretty clear that the diode D2 should be able to handle 100 watts continuous without melting down.

Now, the other diode may be a problem, since it only has 5 volts of reverse bias across it! With similar reasoning, P=V²/Z, so V=√(P*Z)=
√(100*50)=70.7 vrms = 200 volts peak to peak with only 5 volts of reverse bias across it.
In other words, maybe the RF signal is actually being rectified in the "reverse biased" diodes, and turning itself on (appearing to not be able to handle the high power during lab testing).

 

[kspalla]

I guess the blanking is done at gate only and hence 100W gate is probably 10W if the gain of 100W is near 10dB

 

[aot]

biff44 - thanks for your help with this.

So it looks like it's best if I go back to the design I used with the 1W gates (which similar to how you described), but with a larger power supply (+-50V) so the diodes don't accidentally turn on when reverse biased.

 

[biff44]

Any high power PIN diode switch has the following ground rules:
Inductors have to handle large currents, so big wire diameter (no 0402 chips here!)
Capacitors have to handle large RF voltages.
You generally avoid causing resonant circuits, as the voltages and currents are much higher.
You need to get the heat out of the components, espcially the PIN diodes.
You need to have a big enough forward DC current to make sure the Rs of the diode is low.
You need to have a big enough reverse bias voltage so the diode does not conduct. The higher the frequency, the lower that reverse bias has to be (called punch-through) due to bulk silicon properties.
You want to avoid switching the PIN diode state when full power is applied. If that is a requirement, you have to way over-design the thermal part of the design.
You want to test, with expected worse case load vswr, the whole thing, and improve the design where it is melting/arcing over.

Good luck.

Rich


p.s., scope this out, starting at page 11:
www.qsl.net/n9zia/pdf/pin_diode_handbook.pdf

 

[kaggie]

Just looking at the schematic, I can think of a number of things. You'll have to excuse me for not having read anything of the follow ups, and I am really busy right now, but I'll answer your questions:

"1. The matching network I mentioned, is that the right way to sort out the parasitic reactances and get a better match, or should I be series resonating/shunting them out first and then build a matching network if necessary?"

Yes. A 10 W 50 Ohm resistor would work for testing purposes.

"2. is this power non-linearity I'm seeing due to the diodes not being reverse biased enough, or is it just due to the design and I should try to account for it when building the matching network."

Yes to the first. Check your diodes - they are probably not working after you've used them once or twice.

"3. Is this even a good way to build a RF switch, I've had a look on the net and seen some TR switches, many are the series/shunt type T/R switches, perhaps I should be going with those - power handling was my concern though."

A TR switch can handle a lot of power though - that's the whole point of a TR switch. A TR switch is used to transmit to say a body coil which can have CW power that is very high (900V, 50 ohm), and yet, it is also used to receive very small signals and protect the circuitry of the preamps. They transmit to coils with a lot of power.. I highly recommend starting with that. This could work, but as mentioned above, there are a number of problems to it. If you want, you can send me a message, although I may not be able to help you for a couple of weeks (middle of summer.. very busy with trips, work, etc.).

 

[aot]

biff44 and kaggie, thank you for your help here. I had forgotten about the PIN diode handbook, although in the past I have used it mainly for the MRI chapter...

Taking into consideration all that has been said here I have redesigned the gate. Given that when I don't want to transmit the amplifier is blanked and putting out only a small amount of noise (less than 5mV), I have decided to remove the arm of the circuit that shunted the RF to a 50R load during blanking, so the circuit is now just a SPST switch. When the PIN is forward biased it conducts RF and when it is reverse biased it is part of a parallel resonant trap. I also have a C-series, L-shunt matching network on the output. The diode is biased using a +-15V power supply (unfortunately anything higher is unavailable to me as I have no more funding to buy it, nor the time/expertise to build one, and the MOSFETS I am using can only be used up to 30V source-drain voltage), with 100mA of current when forward biased. This gets the resistance of the diode down to about 0.06ohms, and when reverse biased the capacitance is about 3.7pF. I'm getting 9.5k ohms blocking impedance (about 40dB isolation), and in transmit, with the matching network the insertion loss is about 0.15dB. I have tested it with the amplifier, which was fine for up to 100W CW for 60s, higher than that and it goes into 'overdrive' mode and shuts down.