HL2 Digital Bias Problem

[Steve Haynal]

Hi Group,

The digital PA bias, U14 and U15 in the schematic, is not working as planned. The intent was that U14 provides a regulated low current 5V with on/off capability and then U15, a digital potentiometer, divides this down to the correct bias for Q3 and Q4. 3.3V I2C from the FPGA sets the divide ratio. The problem is that the positive side of the resistor divider network (currently 5V) should be the same as or at most 0.3V more than the IC supply voltage (currently 3.3V for I2C to FPGA) but it is not. This causes excess current to flow in the LDO and triggers its thermal shutdown. Stew will not see this yet as U14 is not enabled in any firmware I have distributed. 5V was picked to support Mitsubishi RD16HHF1-like devices which require a bias in the 4.5V range. The current changes I have considered or am considering are:

A. Switch U14 to a 3.3V or 3.5V version. This will allow me to continue testing the PA by just swapping out U14. For the AFT05MS003 devices, we should be able to hit the ~2.9V bias with this. This is also even more similar to the KX3 digital bias setup. I have tested U15 by using the existing not switchable 3.3V as input and can set the divider output from .02V to 3.24V. This is with no load. Unfortunately, this solution does not support the bias voltages expected for 12V LDMOS RF transistors. 

B. Run U15 at 5V but still use 3.3V I2C. This would provide a wider range of bias voltages, but unfortunately the spec requires I2C "high" to be 0.7 of the power supply. For the 3.3V FPGA signals, this means the power supply could be in the 4.5-4.7V range, but that is very close to where we need the bias. Also, this would require an additional switch mechanism for the bias voltage as I don't want to switch off both U14 and U15. I don't like this option.

I am still considering options. Please share ideas. The current fallback is to go with option A, and add an option of using manual potentiometers if a builder wants to use a 12V device on future PCBs. Also, in the next PCB revision, I may switch to the 5K version of U15 and add a resistor so that the steps go from Vdd/2 to Vdd, rather than 0 to Vdd, as that will be more aligned with expected bias voltages and provide smaller steps.

This is an example of the types of problems and changes I expect with the beta2 boards. Expect a few more. For US builders, it is easy to order additional parts. I just ordered new versions of U14 this evening for a few dollars shipping. For international builders, it may be difficult to deal with changing part values. I'd recommend any international builders of the beta2 board build only the minimal nonPA version BOM first. Once the details of the additional circuits are worked out, they can then decide whether to order and add those components.

73,

Steve

KF7O

 

[Dani EA4GPZ]

El 30/01/17 a las 08:11, Steve Haynal escribió:

> I am still considering options. Please share ideas. The current fallback
> is to go with option A, and add an option of using manual potentiometers
> if a builder wants to use a 12V device on future PCBs. Also, in the next
> PCB revision, I may switch to the 5K version of U15 and add a resistor
> so that the steps go from Vdd/2 to Vdd, rather than 0 to Vdd, as that
> will be more aligned with expected bias voltages and provide smaller steps.

Hi Steve,

It's not easy to bodge this into the current board, but for the next
revision I would also consider using level translation of the I2C bus at
U15. This can be accomplished with 2 FETs and a few resistors and it
would allow U15 to run at whatever voltage is needed (say 5V).

73,

Dani.

[Anthony Loveday]

Hi Steve,

You could possibly run U15 at 5V and use a Texas Instruments TXS0102 level translator on the I2C with a couple of extra pull up resistors.

http://www.ti.com/lit/ds/sces640f/sces640f.pdf

Obviously this would require changes to the board with the next revision and I'm not sure how the space would work out. It also won't help with your testing in the short term (unless you kludge something up!!) but you should be fine at present with your option A of swapping U14 to the lower voltage version and using the AFT05MS003 devices.

Thanks to you and everyone else for all the excellent work on the Hermes Lite - this is a great design.

Regards,

Tony.

VK4KRC.

[Steve Haynal]

Hi Dani and Toni,

Thanks for the input. I've used level translators in the past and would prefer to avoid one if possible.

The microchip family of digital potentiometers I am currently using can also be configured as rheostats. Now I'm thinking of a 3 resistor divider network with a digital 0 to 5K sy the bottom. The resistors would be chosen so that the input to the IC's resistors never goes above 3.3V. The tap point for the bias would be 1 resistor away from the top and vary over the range of interest for the particular PA transistor, for example 4 to 5V, or 2.5 to 3.5V. The top of the network would still be driven by a switchable LP2985 LDO. This would have to be a higher voltage, but Digikey has them up to 6.1V. Another variant may go up into the 7V range. The voltage steps won't be linear, but that doesn't matter much for our application. Currents will vary a bit too. Do people with more experience biasing PAs see any problems?

73,

Steve

KF7O

[in3otd]

Hello Steve,
I was thinking of using a divider as you described too, but some first calculations do not give reasonable resistor values for the higher Vgs devices - will try some more combinations.
You can use also a couple of SOT-23 BSS138 as level translators if you want to use a 5 V device; I've used them for an SD card and worked well up to a few MHz. But IIRC 5 V was anyway at the limit for biasing the RD16HHF1.

Smaller Vgs steps will be helpful, IIRC the devices we use have a transconductance between 0.5 S and 1 S, so without a divider one LSB at 3.3 V could be more than 10 mA; maybe not a big issue but it means we will actually use very few steps.
Adding a double opamp like a dual LM358 may make things simpler but it will be one more part to fit on the board.

73 de Claudio, DK1CG / IN3OTD

[in3otd]

Hello again,
here is a divider circuit as you described, with component values for a bias range of about 2.6 V to 3.5 V - should be ok for the AFT05MS003

The red dashed line in the top graph is at 3.3 V, the thin red line is the digital rheostat high side value (low side is the green one) and the bias is the thin blue line.
A similar circuit should work for the RD15HVF1, which has also a low threshold Vgs, with low dispersion, while the RD16HHF1 will require a higher voltage source and, according to the datasheet, can have a much larger threshold range.

73 de Claudio, IN3OTD / DK1CG

[Steve Haynal]

Hi Claudio,

Thanks for the design. I see that R3 is necessary for a small window with the AFT05MS003. I also have another variation where R3 is 0, V1 6.1V and bias sweeps from 4V to 5V. (Left the R4 and R2 values on a paper at another place...) This design is general enough that builders have the option for various bias sweep ranges.

73,

Steve

KF7O

[in3otd]

Hello Steve,
actually R3 is not really needed, is there mostly because of some previous experiments. You can set R3 to zero and add its former value to R2, the circuit will work the same. Not sure R3 will actually be useful for other devices with a different voltage range, if we keep the same kind of network.
A 6.1 V LDO could be used, I'm still wondering if will cover also the wide Vth of the RD16HHF1. I saw other LDOs, with a different package are often available with voltages up to 8.2 V, maybe we could use on of those, with 5 V for the AFT05MS003 and RD15HVF1 and the 8.2V for the RD16HHF1, if we want to cover that also.

[Steve Haynal]

Hi Claudio,

For the RD16HHF1, the amps from Hobby PCB and the one John built all use a 5V regulator with POT for the bias. My target of 4-5V comes from these designs.

73,

Steve

KF7O

[John Williams]

The Regulator is variable and is configured for 6V IIRC. 5V was not quite enough. Glenn suggested the change while we were testing the amp.

John

--
You received this message because you are subscribed to the Google Groups "Hermes-Lite" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hermes-lite...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Graeme Jury]

Hello Steve,

If running the RD16HHF1 at high power (20 watts+) it needs to be biased more into class ab2 with ~200 mA or more of drain current. To get this 6 volts is needed but for more modest powers the output is clean with the amp biased more towards class B and 5 volts is enough for this.

73, Graeme ZL2APV

--

[Graeme Jury]

Sorry I was talking about a pp pair but the bias still applies to running a single amp in class A with 5 or 6 watts out. They need heavy bias around 500 mA or so for best linearity in class A single ended.

Graeme

[Steve Haynal]

Hi Group,

Thanks for the feedback. For the 4-5V, I was looking at John's original schematic and the current Hobby PCB RS-HFIQ. The LP2985 LDO can be found at 6.1V, but with the proposed voltage divider network it will be difficult to produce bias voltages close to 6.1V. The opamp RF preamp requires a well regulated voltage in the 9 to 10 V range. I am now thinking of just using that at the top of the resistor divider network with a mosfet to switch the voltage to the divider network on and off. The mosfet would take the space of the LP2985. Four more resistors would be added. Builders would pick resistor values to place their bias window where they want.

The HL2, including use of the enclosure for heat sinking and PA biasing, did not really have a 20W amp in mind, so not really sure that much power will ever work out.

73,

Steve

KF7O

[Graeme Jury]

Hi Steve,

I had just been working on a 20 watt PP RD16HHF1 amp and had it on the brain. Yes it won't go in the enclosure and was to be an external amp driven by the HL but the bias voltages were still relevant. It seems that the RD16HHF1's are not prone to thermal runaway and your bias system should work well. I spent a lot of time messing around with thermal compensation and basically wasted my time as it worked just the same with or without it.

73, Graeme ZL2APV

[in3otd]

Hello,

good idea to use the 10 V supply of the driver; here are the circuits with a 10 V source, for the AFT05MS003 and RD16HHF1. Seems we should be able to cover both.

73 de Claudio, IN3OTD / DK1CG

[Steve Haynal]

Hi Claudio,

I'm having second thoughts about using the 9 to 10 V driver supply for the bias too. With the AFT05MS003, both the driver and the PA use the same higher current 9.44 V supply. I am worried that the varying PA loading will cause adverse fluctuations in the bias voltage. Maybe this feedback will actually be good. What are your thoughts? For 12V RF LDMOS devices, there is a separate low current 10V supply for the driver and I am not as concerned about sharing this with the bias.

 

73,

Steve

KF7O

[in3otd]

Hello Steve,
using the values in the latest circuit for the low-voltage PA above, the gate bias line has an equivalent source resistance of 8 to 9 kohm, depending on the actual rheostat value; with the 2x100 nF currently on the bias line this makes a low-pass filter with a corner frequency below 100 Hz, so any fluctuation above this frequency will be attenuated. The DC/DC datasheet does not show typical transient responses, but I'll expect that slow transients will be well filtered. In the worst case we can increase the gate bias bypass caps.
DC output changes due to increased current from the PA when this is driven will be seen, attenuated, by the gate bias but I'll expect these to be quite small. I guess the only way to find how well it works is to test all this in practice, hi.

73 de Claudio, IN3OTD / DK1CG

[Gilbert Cross]

Hi, group. Having trouble with the part number for U1 in the
BOM for V2 beta, not recognized by either Mouser or Digikey.

Thanks Gil K8EAG

[in3otd]

Hello,
?, Octopart says it's available from both, https://octopart.com/search?q=S25FL116K0XMFI041++

73 de Claudio, IN3OTD / DK1CG

[Dani EA4GPZ]

El 02/02/17 a las 17:11, Gilbert Cross escribió:

> Hi, group. Having trouble with the part number for U1 in the BOM
> for V2 beta, not recognized by either Mouser or Digikey.

Hi Gilbert,

Perhaps this direct link helps:

http://www.digikey.es/scripts/DkSearch/dksus.dll?Detail&itemSeq=217657837&uq=636216410856819388

73,

Dani EA4GPZ

[Steve Haynal]

Hi Group,

Just a quick update. The main T/R relay was not working for me. There is an error on the PCB as the footprint is a mirror of what it should be. I interpreted the datasheet as top down view when I should have thought of it as bottom up view. Fortunately only two pins are affected and the relay can be mounted at a slight angle and two short jumpers run to those two pins. I will start a list of all mods next week on the wiki.

I received and installed my 3.3V LP2985 regulator. Now the digital POT is working at least for bias values in the range required for the AFT devices. I attempted to set the bias but smoked one of my transistors. I was making silly mistakes of having no proper load and not starting with the lowest possible bias voltage. I also have no current limited power supply. It is also hard to measure the bias current with the current board and I may add some facility to make this easier on the next board. I am gearing up to try again tomorrow evening.

Frying a transistor worried me about running this PA at 9.44V. Claudio's testing was at 9V, but I was trying to pick a value between 9 and 10V as the RF driver was designed for 10V. I adjusted the 9.44 regulator to produce just under 9V. I tested the low power RF output and it appears to be fine, in the 19 to 20 dBm range. I think I will keep this at 9V and add this mod to the wiki. We may need to fine tune a few components values on the driver.

73,

Steve

KF7O

[in3otd]

Hello Steve,
to limit the current you can also put a series resistor on the output transformer center tap, say about 5 or 10 ohm. Should be small enough to allow setting the bias properly and big enough to limit the max current to safe levels. Decrease to 1 ohm for the first high-power tests before removing it completely, hi.
Even Vdd=10 V should be safe for DC biasing, the device has 12.5 V as maximum operating voltage and 30 V as maximum Vds. The problem of running it at higher voltages are the peak RF voltages that can be present with a mismatched load. With 3.3 V on the gate I'll guesstimate that the drain current could be in the 500 mA range, still a safe value if kept for a short time and with some heatsinking, maybe your PA was oscillating due to the lack of a proper load.
I too wondered if there should be a dedicated current sensing for the PA but I don't know if there is enough space/resources for that.

73 de Claudio, IN3OTD / DK1CG

[Glenn P]

There's a handy little IC that can be used to measure high side current, with a small sense resistor. Its in a tiny SOT23-5  package.   INA139NA/250

On Friday, February 3, 2017 at 6:26:53 PM UTC+11, Steve Haynal wrote:

Hi Group,

>>>>>>>>>
 

I received and installed my 3.3V LP2985 regulator. Now the digital POT is working at least for bias values in the range required for the AFT devices. I attempted to set the bias but smoked one of my transistors. I was making silly mistakes of having no proper load and not starting with the lowest possible bias voltage. I also have no current limited power supply. It is also hard to measure the bias current with the current board and I may add some facility to make this easier on the next board. I am gearing up to try again tomorrow evening.

[Steve Haynal]

Hi Glenn,

Thanks for the link.  There are currently 4 channels in the slow op amp/adc circuit: 1 FWD PWR, 1 REV PWR and 2 PA thermal. I'm thinking that we really only need 1 PA thermal and I can reuse one op amp/adc channel for current sensing. One option is to use a current monitor like you suggest. Another option is to switch to an op amp with rail-to-rail or beyond-rail inputs and use an op amp input directly for measuring the differential across the sense resistor. Still another option is to use an INA219 which has its own I2C interface and can be connected to the existing I2C bus:

https://www.adafruit.com/product/904

I expect changes in the slow ADC op amp/adc circuit as it has never been tested. Once I (or others) test that, we can decide on a final solution that includes current sensing. I would really like the Hermes-Lite 2.0 to be able and measure the current to the PA.

73,

Steve

KF7O

[Steve Haynal]

Hi Claudio and Group,

My concern with 9+ V is for peak RF voltages. I had no scope connected but think the PA could have been oscillating and producing RF. I feel safer at 9V, especially if the driver will work well at this voltage. The datasheet uses 9V to specify SWR ruggedness. The AFT transistors would be difficult to replace with only a soldering iron. I have been using hot air. I don't want people to have to replace these frequently. 

I added a 5 Ohm current limiting resistor, proper load and better heat sinking. I am no longer smoking transistors, but I suspect something is still wrong. For the transistor I replaced, I can set the bias to a pretty steady 100mA when the bias voltage is ~2.36V, lower than what you reported. For the other original transistor that did not smoke, I don't see 100mA until its bias voltage is ~3.00V, closer to what you saw. I see more current drift with this device though. Also, it was oscillating a bit at first but that settled down once I soldered its leads better. I never felt much heat from either device and actually thought everything was cooler than anticipated, even with the minor oscillation. This bias voltage differences seem too large a variation to me so I did not try any full power tests. I suspect that the original transistor may not have smoked but may still have been damaged in my earlier test. I also want to take a close look at all my soldering of PA components. I ran out of time and am traveling from today until next weekend, so this will have to wait until next weekend. Any other thoughts or suggestions?

73,

Steve

KF7O

73,

Steve

KF7O

[in3otd]

Hello Steve,
I agree that the PA may have been oscillating and produced too high peak voltages - should not happen with a stable PA and proper load but transmitting without (a proper) antenna may always happen so it would be good to have a PA that can withstand this condition.
My devices are likely from the same batch as they came from a single order, in the same strip; their Vgs for 100 mA bias is practically identical.
I can try to see if it's possible to rewire the transformers connections to use just one-half of the PA at a time, for checking the devices, but I'm not sure this will take less time than changing also the other device, just to be on the safe side.
You can also check if the drain current goes back to zero when lowering again the bias voltage; if they are oscillating they sometimes go into some "self-biasing" mode (seen this sometimes with the bigger Mitsubishi devices)

73 de Claudio, IN3OTD / DK1CG

[Steve Haynal]

Hi Group,

I am designing the current sense circuit and am having trouble achieving good accuracy. Assuming 0.1 Ohm sense resistor, this produces a voltage of 10mV for 100mA of bias. But at low levels of 10mV, devices such as the INA139 do not have great accuracy. See table 2 where error is 5% to 10% which means current measurements at this low level can be off by 10 mA. This is not taking into account errors from resistor precision and the follower op amp. I guess the *relative* current measurements of the two LDMOS devices will be nicely in sync. I don't want to increase the sense resistor as at full PA power it will be burning ~200 mW and see a drop of .1 to .2 V.

There is another device with builtin ADC and IC2 interface that appears to offer better accuracy: 

https://www.adafruit.com/product/904

http://www.ti.com/lit/ds/symlink/ina219.pdf

I'm not sure I believe it for low currents.

Any ideas or thoughts on achieving better accuracy? Or is the accuracy of the INA139 about what is expected? 

73,

Steve

KF7O

[Graeme Jury]

Hello Steve,

The ZXCT1021 looks pretty good in the 10mV range. They are $2 in one offs at Digikey.

73, Graeme zl2apv

[in3otd]

Hello,
I think that the accuracy at low currents is mainly limited by the device input offset voltage.
For the INA139 the offset spec is ±0.2 mV typical, worst case is ±1.5 mV, which would give 15 % error at 10 mV of sense voltage.
The INA219 has much better specs: the "B" version at lowest PGA gain has ±50 uV worst case, the "A" version at highest PGA gain has ±200 uV, worst case, which is still only 2% at 10 mV of sense voltage.
The ZXCT1021 specs are less clear, but from the graph on page 4 of the datasheet the offset appears to be around ±0.5 mV (worst case ?) but the table on page 2 seems to imply almost ±0.9 mV at 30 mV of Vsense.

If the current sensing is just for the PA I think one could just measure the offset without any bias and subtract that from the measurements to improve the accuracy.

73 de Claudio, IN3OTD / DK1CG

[Takashi K]

Hi,

How about ON semi NCS210 ?

It's  low offset voltage ±60uV(Max). but there is no graph on the datasheet.

73, Taka  ji1udd 

[Steve Haynal]

Hi Taka,

Thanks for the NCS210 link. It looks like a very promising device. The specs look good for low Vsense, but I'm not sure of how to deal with the fixed 200x gain during full power. Maybe we just let the device go full scale during full power TX and only use the current measurement for low currents?

73,

Steve

KF7O

[Steve Haynal]

Hi Graeme,

Thanks for the link. The device is available on AliExpress in the $1 range. I was looking at a similar device:

https://www.diodes.com/assets/Datasheets/ZXCT1107_10.pdf

which has a nice design example taking into account error starting on page 11. The ZXCT1021 appears to have 3% error in the 10mV Vsense range which is better than the 10% or 18% of the devices I was looking at. 

73,

Steve

KF7O

[Steve Haynal]

Hi Claudio,

Your suggestion works best if the voltage-current converter in the device is linear below 10mV. Some datasheets I've read only guarantee linearity above 10mV. See the true offset comment on page 8 of this. Still, the correction is better than without correction. How much do you estimate accuracy can be improved?

I am also thinking about just a voltage divider after the sense resistor that feeds one of the LM324 opamp channels. The voltage variation from the sense resistor would be divided down to some where within the range of the opamp. The opamp would then provide gain so that what feeds the ADC channel still sweeps across enough range to provide resolution of ~1mA steps. This is a rough idea and I haven't yet worked out any details. Various offsets and introduced nonlinearities could be corrected in software. The BOM cost would be reduced. What do you and others think?

73,

Steve

KF7O

[Steve Haynal]

Hi Group,

Thinking a little bit more about direct use of a LM324 channel. With a 0.1 Ohm sense resistor, each milliamp is 100uV delta. If voltage dividers (sort of a bridge) are used to reduce the voltage to the range of the LM324 driven with a 3.3V supply, then this goes down to around 35uV per milliamp. I'm currently assuming the LM324 can't be driven with a higher voltage as both of the higher voltage supplies are switched on and off at times. It needs to be on all the time for other measurements. It should be possible to measure and remove the input offset voltage, but the LM324's Vos drift of 7 to 20 uV per degree C would be unacceptable. But perhaps the 2 uV/C input offset drift of the TSV324A would be acceptable.

Still thinking about various options here.

73,

Steve

KF7O

[in3otd]

Hello,
I also thought a little about using just an opamp to sense the voltage drop but there are many possible source of errors and I'm not sure on how to take all of them into account. I also thought to use a rail-to-rail input opamp powered by the 12 V supply to avoid the input divider but of course these are a little more expensive than the one currently used on the H-Lv2b2; the cheapest I could find  are the TLV2374 and TS924. A nice circuit whose CMRR does not depend on resistor matching is described here and it seems it's the same circuit implemented in some of the current-sensing amplifiers we are currently considering.

We may get a somewhat improved accuracy from one of those devices if we just measure the current difference when switching on and off one of the LDMOS to avoid the lowest current region, something like this:
- set the bias of one LDMOS to get around 100 mA; it could actually be from, let's say, 80 mA to 120 mA
- leave the first LDMOS on and  set the bias of the second one to get a reading of 100 mA more; now this current difference should be closer to 100 mA, since the current-sensing amplifier is operating away from the  zero voltage point.
- repeat from the beginning, swapping the devices.
Things will actually be a little more complex due to the thermal drift of the bias current but maybe with several readings with the LDMOS on and off we could compensate/estimate this.

73 de Claudio, IN3OTD / DK1CG

[Takashi K]

Hi Steve,

Voltage divider is placed btw NCS210 and ADC.

NCS210 costs $1.32 at Arrow Electronics.

73, Taka ji1udd

[Steve Haynal]

Hi Taka and Group,

I took a closer look at the NCS210. It turns out that there is a similar part, NCS199 or INA199, that is more widely available with lower fixed gain of x50 for the x1 suffix parts. See: http://www.ti.com/lit/ds/symlink/ina199.pdf. This part costs around $0.88 in single quantities from DigiKey or Mouser. It shows up in searches of www.aliexpress.com. I am strongly considering this part with a 40 to 50 mOhm sense resistor and no voltage divider on the output, just feed the follower LM324 op amp channel for buffering. This is similar to the typical application in the datasheet. There is a similar part, INA213 or NCS213 but it is hard to find and doesn't have any better specs.

73,

Steve

KF7O

[Takashi K]

Hi Steve,

INA199 seems good !

73, Taka ji1udd