MMIC amplifiers power up sequence?

[CopperCone]

So these guys have a complicated power up sequence. You need to sequence a few voltages, then adjust a voltage to a specific current.

This is kind of a pain in the ass as far as overhead goes. Are these steps absolutely necessary?

I would like to find the correct voltage the bias current is right, then turn everything on at the same time... otherwise it means fairly complicated circuitry or manual adjustment every time.

The fact that it has a power down sequence makes it even more complicated.

Easy to implement with a few specialty IC's.. but jeeze

can some kind of latchup happen? (unavoidable due to the fact that the high frequency nature of the device means that anti latchup circuits are not implemented)

[kg4arn]

I won't know the answer but I am curious as to which parts/mfg are you referring to.

[CopperCone]

Analog parts do it. The microwave LNA and power amplifiers.

[rfeecs]

The sequence is usually to have stable gate voltage before drain voltage is applied, and gate voltage stable after drain voltage is removed.

This is mainly to avoid getting into a region where the devices oscillate.  Also, they are depletion mode devices.  So applying the drain voltage first would cause excessive current until the gate voltage is applied.  Either way, there is a danger of destroying or damaging the devices.

Typically you have to manually adjust the (negative) gate voltage one time, possibly tweaking for each unit.  Then it should be repeatable.

This is also a problem for large GaAs FET based discrete power amps.  At a minimum, people use a simple sequencer with some RC time constants built in so the drain voltage application/removal lags the gate voltage.

[CopperCone]

how repeatable is the gate voltage? I just designed a complicated circuit to set it each time *but not a servo loop*. took me like 3 hours  for just the circuit schematic

If I build a power supply, should I bother? It increments a DAC with a counter until the voltage is reached, then a comparator turns it off (so there is no MCU with code on it that needs to be replaced if it breaks). Also, no MCU noise to mess with stuff.. (never measured how well isolators and stuff work for the very high frequency noise a MCU makes.. plus I don't know what the spread is on their noise generation).

[rfeecs]

Like I said,

Typically you have to manually adjust the (negative) gate voltage one time, possibly tweaking for each unit.  Then it should be repeatable.

How repeatable?  Depends on your requirement, I guess.  GaN should be less sensitive than a super low noise pHEMT, just because GaN usually has a pretty large pinch-off voltage.  But things will drift with aging and vary with temperature.  I would guess the current should stay within 10% of it's initial setting.

Some power parts may require significant gate current (could be either positive or negative) when they are driven hard.  That's another thing to keep in mind with the negative supply for power MMICs.

[CopperCone]

yes, it certainly appears you need alot of support equipment. 10% sounds pretty bad, may as well build the circuit. to circumvent aging. i thought a auto zero might make fake modulation that could be confusing, compared to thermal drift, which is smooth.

perhaps its better to switch it to a constant current source.

Currently I am using very low noise parts for this, LT3090 and LT3045. This is for a preamp.

It would be better to swap the LT3090 for a constant current source, that's initially set to something with a resistor to make it act as a voltage source, then to remove the voltage set resistor and let it go into constant current mode, but the noise on a current source would probably be substantially higher. I am not sure what to do. I guess I should go for thermal stability.

I see the LT3045 datasheet has a circuit for a low noise current source, specifically for RF bias, I guess it should be possible to make one out of the LT3090.

[rfeecs]

It would be better to swap the LT3090 for a constant current source, that's initially set to something with a resistor to make it act as a voltage source, then to remove the voltage set resistor and let it go into constant current mode, but the noise on a current source would probably be substantially higher. I am not sure what to do. I guess I should go for thermal stability.

I see the LT3045 datasheet has a circuit for a low noise current source, specifically for RF bias, I guess it should be possible to make one out of the LT3090.

I have no idea what you are talking about.  I'm confused.  What MMIC are you trying to bias?

[CopperCone]

Sorry. The question should be, is there any benefit towards starting with a constant voltage source to satisfy initial datasheet conditions, then switching to a constant current source, or should an adjustment be made and then the source remain constant voltage. '

I'm not sure which part I will be using, but they typically state something along the lines of setting a rf bias voltage, then setting one or two more voltages (this can all be done using a sequencer), then adjusting the RF bias current to a specific level. I thought it was across the board, because I saw this with a 10MHz-10GHz LNA preamp, a 1GHz-11GHz preamp and several chips capable of putting out 25-40 watts of energy. (the higher power devices are even more complicated, as multiple voltages seem to transition from constant current to constant voltage according to the bias procedure, not just one like with the LNA). I do however think that if I get one of the higher current devices I will manually set it using lab supplies, at least for now, since making a multi-amp dc power system that's also low noise and small is not a trivial task.

This part looks like a worthy candidate if you want specifics (but this seems to apply across the board):
http://www.analog.com/en/products/amplifiers/rf-amplifiers/hmc753.html#product-overview

For the last step I can either:

1) Add some fixed offset to the voltage to set it at a particular voltage that corresponds to the correct current draw specified by the datasheet and leave it alone. This is easy, and I feel confident that my DC power system will be reliable. Now I took this a step further by measuring the initial conditions each time with my 'one shot' feedback loop, but it only occurs once to counter the effect of aging.

2) Switch the voltage source to a constant current source which is set to the datasheet current (or use a servo loop involving a ADC and a DAC). This will require investigation as I am not sure about what low noise power supply topologies are OK with being hot swapped into working voltages and not oscillating etc.

The datasheets I have seen are unclear about this, and a sales representative just told me to get the demo board (which says absolutely nothing about the biasing system).

I'm not sure what will happen if there is some kind of feedback on the bias current, while its transmitting, I thought it might oscillate corresponding to modulation and create some kind of unwanted bi-products. I would like to eliminate thermal drift... but I don't want to approach that challenge if it causes additional problems. Normally I would just go for a constant current source, but I don't know much about the technologies used on these advanced chips.

[rfeecs]

Sorry. The question should be, is there any benefit towards starting with a constant voltage source to satisfy initial datasheet conditions, then switching to a constant current source, or should an adjustment be made and then the source remain constant voltage. '

I'm not sure which part I will be using, but they typically state something along the lines of setting a rf bias voltage, then setting one or two more voltages (this can all be done using a sequencer), then adjusting the RF bias current to a specific level. I thought it was across the board, because I saw this with a 10MHz-10GHz LNA preamp, a 1GHz-11GHz preamp and several chips capable of putting out 25-40 watts of energy. (the higher power devices are even more complicated, as multiple voltages seem to transition from constant current to constant voltage according to the bias procedure, not just one like with the LNA). I do however think that if I get one of the higher current devices I will manually set it using lab supplies, at least for now, since making a multi-amp dc power system that's also low noise and small is not a trivial task.

This part looks like a worthy candidate if you want specifics (but this seems to apply across the board):
http://www.analog.com/en/products/amplifiers/rf-amplifiers/hmc753.html#product-overview

For some reason your message didn't render properly, part of it was missing.
There is no transition from constant current to constant voltage.  Using a constant current source will not work.  All the sources should be voltage sources.

The HMC753 does give an elaborate biasing procedure.  I think it's a bit overkill:

The recommended biasing procedure during power-up is as
follows:
1. Connect GND.
2. Set VGG1 to ?1 V.
3. Set VDD to 5 V.
4. Set VGG2 to 1.5 V.
5. Increase VGG1 to achieve a typical quiescent current (IDQ) =
55 mA.
6. Apply the RF signal.
The recommended biasing procedure during power-down is as
follows:
1. Turn off the RF signal.
2. Decrease VGG1 to ?1 V to achieve IDQ = 0 mA.
3. Decrease VGG2 to 0 V.
4. Decrease VDD to 0 V.
5. Increase VGG1 to 0 V.

 

You just want to keep Vg2 below Vcc.  You could just bias Vg2 with a voltage divider from Vdd.  Then all you need to do is turn Vg1 on, then Vdd on.  Turn off Vdd, then Vg1.

I wouldn't get too elaborate.  Just tweak Vg1 one time and leave it.  Things should be stable enough.  Don't worry excessively about the noise from the power supplies.  That low frequency noise won't affect the noise figure up in the GHz range.  It may affect the close in phase noise a bit, but I'm assuming you are not worried about that.

I would keep the bias as simple as possible to begin with.

[CopperCone]

ah, good call on the current source not working.

Well I am not really worried about close in phase noise, but given that the part itself costs 90$, I might as well pay a little extra IMO. I wanted to cascade two of them, my only worry is that nearby signals might saturate such an amplifier.

Since I want to cascade them, I figure paying a little extra for my anti-aging scheme is worth it, as the cost is better absorbed.

I made the circuit using a LT3090, LT3045 , LT1014 (have some), 74HC590 (counter), LTC6993 (oscillator, have some), LTC2924 (sequencer), AD7801 (have some).. parts been layin around for years so why not.

I want to etch the board then refuse an old RF can from a defunct spectrum analyzer to house the amplifiers. I was going to deadbug them upside down next to each other and drill holes in the RF can to stick hardline into them, (rg141 or smaller), then solder the hardline directly to the input pins. Very curious to how it will work, I also salvaged some RF feedthroughs.. call it an experiment in guerilla microwave design. I suspect the RG036 I have would work very well, unfortunately the SMA adapters for it are 30$ each... gonna have to use the standard sizes.

For the grounding of the packages I was gonna lay em upside down on a piece of copper and put a strip of copper that's the same dimensions as the center pad across it, then have it branch out and solder it down (kinda like a bowtie pasta shape ground overpass).

I'm guessing that a RF PCB would cost like >100$.. do they sell fixed dielectric constant ceramic PCBs that have photo resist on them for etching? I also wonder if carbide would work on drilling holes in these, and if a copperset via kit would crack it from the anvil and via deformation.

and yes, I like working on tiny things

[rfeecs]

Well good luck with the guerrilla microwave.  I don't think it will work.  Particularly the grounding doesn't sound good.  But if I were spending that kind of money on MMICs, I would get a proper board made at least.  You really need a proper housing as well if you are going up to 10GHz.  But yes, it all costs money.

[CopperCone]

I think you understand exactly what I meant, but, just incase

I wonder when it would start to misbehave, I just can't get used to the idea of an sheet induct or that is ~1.65mm x 3.8 mm (per side) mattering.. I know it does but damn, to me that's a standard short lol

I calculated the inductance to be ~328pH on each side, so that halves (?) to 164pH, which at 10GHz is 10Ohm of reactance.

I guess that's pretty bad.. out of curiosity what is considered acceptable here?

I could see a better grounding scheme possibly be to make a copper sheet with cutouts so that the voltage and RF pins of the device are hanging over a little void so the soldering could be done and wires routed to the sides that need them. I think that would be almost (zero?) inductance between the ground pad and the ground plane then.

like this

I wonder how to cut tiny trapazolidal holes in copper without a laser though. I guess a tiny dremel endmill

[rfeecs]

Think about the path the ground current starting at the coax tip has to follow.  You are creating a huge loop.
Also, part of the loop is common to input and output, causing a lot of feedback.

You can give it a try, I guess you never know.

[CopperCone]

I actually found aluminum nitride PCB selling for ~50 dollars. I think you can toner transfer etch them. I might do that. I thought it would be 500$.

also drillable. i think

[KJDS]

There's an Ampleon application note about a biasing circuit for GaN power devices which explains a reasonable way of doing things.

For lower power, then I think there's a circuit given on the ATF54143 family of devices datasheet that uses a pnp to bias and it's reasonably easy to calculate the values to go round it.

[ThomasDK]

Why not used the power controller recommended by ADI and call it a day? 

http://www.analog.com/media/en/technical-documentation/application-notes/AN-1363.pdf

[CopperCone]

 :scared:well that sure looks interesting

this whole microwave thing is like the part in men in black when he kicks open the door and its a locker in an alien high school or subway or whatever