RF input voltage protection and failed AD9866

[Dani EA4GPZ]

Hi guys,

It seems the AD9866 on my Hermes-Lite 2.0 beta2 has died. It started
shorting out the 3.3V supply randomly, making the polyfuse trigger and
resetting the whole board.

I've pulled out the chip and ordered a replacement, so I'll only be
certain that the AD9866 was the culprit when the replacement is
installed. However, the rest of the board seems to work fine with the
AD9866 removed.

To be honest, I think this is my fault. The other day I was running some
calibrations in Linrad which involve connecting the RF input to a pulse
generator. You can see more details here:

http://destevez.net/2017/04/calibrating-the-hermes-lite-2-0-beta2-in-linrad/

http://destevez.net/2016/12/testing-a-simple-pulse-generator-for-linrad-calibration/

I'm thinking that the pulses from the generator might have been too
strong for the AD9866, producing some damage. However, it took more than
48 hours after the calibration for the fail to occur. The unit had been
running without problems for many days, so I'm suspicious that these
incidents are connected.

I recall someone posting on this list that the current diode protection
D1 was probably insufficient, since its threshold voltage was too high
and given the transformer action of T2, the maximum allowed voltage at
the input of the AD9866 could be exceeded.

My pulse generator produces 4.6Vpp pulses into a 50Ohm load (but the
pulse duration is 10ns), so it's definitely a bad idea to connect the
pulse generator directly to the Hermes-Lite as I ended up doing (I was
using a 10dB attenuator in most of the tests).

In any case, we should probably consider a better replacement for D1
just in case someone is as stupid as myself.


73,

Dani EA4GPZ.

[Daniel Estévez]

El 14/04/17 a las 13:55, Dani EA4GPZ escribió:

> I recall someone posting on this list that the current diode
> protection D1 was probably insufficient, since its threshold voltage
> was too high and given the transformer action of T2, the maximum
> allowed voltage at the input of the AD9866 could be exceeded.

More precisely, it was Steve KF7O, who said

> My concerns with D1 are not pcb related, but to make sure the AD9866
> is protected. Right now I think D1 allows peak to peak voltages of
> 5V. With a 9:1 impedance ratio transformer you'll see 15V at the
> AD9866. This may exceed some maximum rating. (Check the datasheet.)
> It may have worked fine for the Hermes with a 4:1 impedance ratio
> transformer and the Hermes' different ADC, or maybe they didn't even
> consider this a problem.

Currently D1 is still 5V. The datasheet of the AD9866 says that the
maximum voltage of RX+ and RX- is AVDD+0.3V. So this leaves us with a
maximum of 7.2Vpp after T2, which is an 8:1 transformer, so voltage
should be clamped to 0.9Vpp or lower at D1.

I think I've done the calculations correctly, but probably some should
check them.

73,

Dani.

[Dani EA4GPZ]

El 14/04/17 a las 14:11, Daniel Estévez escribió:

> Currently D1 is still 5V. The datasheet of the AD9866 says that the
> maximum voltage of RX+ and RX- is AVDD+0.3V. So this leaves us with a
> maximum of 7.2Vpp after T2, which is an 8:1 transformer, so voltage
> should be clamped to 0.9Vpp or lower at D1.
>
> I think I've done the calculations correctly, but probably some should
> check them.

Silly me! Voltage transformation goes with the square root of the
transformer ratio, so that should be 2.55Vpp instead of 0.9Vpp.

73,

Dani.

[in3otd]

Hello Daniel,
sorry to hear about your damaged AD9866; your calculations on the maximum input voltage are mostly fine, the actual maximum swing is slightly lower since the RX inputs of the AD9866 are internally biased at about 1.3 V - this is briefly discussed in the datasheet, page 33 at the bottom. From the numbers there the maximum peak voltage on the 50 ohm side should be about 1.1 V (so a swing of 2.2 V). I'm not sure if there is a TVS diode with such a low breakdown voltage, we could just use a couple of schottky diodes in series. Nevertheless, 4.6 Vpp into 50 ohm is quite a big amplitude and maybe a simple diode could not be able to clamp it effectively.
Out of curiosity, did you use the the diode in Steve's BOM for D1 and mount it with the polarity as indicated on the PCB silkscreen? Because that device is an unidirectional TVS and should have limited the positive voltage swing to one diode drop but maybe that internal diode was not fast enough to protect the device.

73 de Claudio, IN3OTD / DK1CG

[Daniel Estévez]

El 14/04/17 a las 16:50, in3otd escribió:

> Out of curiosity, did you use the the diode in Steve's BOM for D1 and
> mount it with the polarity as indicated on the PCB silkscreen? Because
> that device is an unidirectional TVS and should have limited the
> positive voltage swing to one diode drop but maybe that internal diode
> was not fast enough to protect the device.

Hi Claudio,

I have the CDSOD323-T05L installed as D1 as indicated in Steve's BOM.
I've double checked the polarity and it's OK (cathode band goes to GND pad).

If I understand correctly the datasheet of this TVS, it will clamp the
voltage to between -0.7V and 5V with respect to ground. Therefore, when
I connected my pulse generator, the TVS would limit negative-going
pulses but not positive going pulses (every other pulse changes polarity).

73,

Dani.

[in3otd]

Hello,

If I understand correctly the datasheet of this TVS, it will clamp the
voltage to between -0.7V and 5V with respect to ground. Therefore, when
I connected my pulse generator, the TVS would limit negative-going
pulses but not positive going pulses (every other pulse changes polarity).

oops, I forgot that you had pulses of both polarities, I just considered only the positive pulse, after looking at the pictures on your blog, hi.
The TVS, mounted with the cathode band to GND, will limit to about +0.7V and -5V with respect to ground. But actually the internal diode seems to have a high threshold, I've just measured mine here with a multimeter (which uses a few mA as test currrent) and the drop is about 1 V.

[Steve Haynal]

Hi Dani,

Sorry to hear about your mishap. Does TX still work?

I've been worried that the RX TVS diode is not a low enough voltage for several years now, but have never found (or looked very hard) for a good replacement. I'd be very happy is someone suggests a better part. Another concern that has been raised is that RX distortion may increase with other (untested and lower voltage) TVS device.

73,

Steve

KF7O

[Dani EA4GPZ]

El 16/04/17 a las 08:14, Steve Haynal escribió:

> Hi Dani,
>
> Sorry to hear about your mishap. Does TX still work?

Hi Steve,

It's not that RX became deaf.

When the AD9866 failed, it started shorting the 3.3V supply to ground
after a few minutes (or seconds) of turning the unit on. This would
cause everything to reset many times rapidly and the FPGA usually
blocked in an unusable state until power was removed.

When power was reapplied, RX was working fine, but the AD9866 started
shorting the supply quickly. I didn't test TX.

Probably the overvoltage made the AD9866 much more prone to latchup.

> I've been worried that the RX TVS diode is not a low enough voltage for
> several years now, but have never found (or looked very hard) for a good
> replacement. I'd be very happy is someone suggests a better part.
> Another concern that has been raised is that RX distortion may increase
> with other (untested and lower voltage) TVS device.

The distortion is a good concern. Probably we just need something that
works well for static protection. We don't need to protect from any
strong RF generator that the user might plug into the input.

73,

Dani.

[Wolfgang DL9UFB]

Hello,

I and many others use for the FA-SDR Transceiver
(www.funkamateur.de) a GBLC03C for the protection
of the receiver input.
The diode is available in the shop www.box73.de.
(Sri, Google Translations)

73 de Wolfgang

[Steve Haynal]

Thanks! Will take a look.

73,

Steve

KF7O

[Dani EA4GPZ]

Hi all,

Just to let you know that I've soldered a new AD9866 and my HermesLite 2
is running perfectly now, so:

* It seems the only problem was with the AD9866. Most likely overvoltage
on the RF input caused the chip to become susceptible to latchup.

* I'm a happy camper again ;-)

73,

Dani.

[Steve Haynal]

Hi Dani,

Great news that your Hermes-Lite2 beta2 is working again. We'll have to see if we can provide better RX frontend protection. Let me know if you have any ideas.

73,

Steve

KF7O

[in3otd]

Hello,

regarding using normal diodes to limit the AD9866 input voltage to safe levels, I did some measurements on the RX IMD after adding some diodes in "antiparallel" after the anti-aliasing filter (on DB16, in practice); the D1 TVS was not mounted.

First with a standard MMBD7000 diode, which contains two diodes in series in a SOT-23 package; I tried with both diodes in series in antiparallel with another two diodes in series and also with just two single diodes in antiparallel.

The IMD performance are unchanged until around 0 dBm input:

I've done the same also with some BAT54 Schottky diodes:

as expected the increased intermodulation due to the diodes appears a little earlier in this case but still above -10 dBm.

I did not measure at which voltage these diode combinations do actually clip the input voltage and I'm not sure how effective these protection will be but at least it seems they should not cause any increase in the IMD for reasonable input levels.

73 de Claudio, IN3OTD / DK1CG

[Graeme Jury]

Hello Claudio,

Thanks for the writeup on the testing. Just to be clear, the second test on the MMBD7000 diode, did that mean that you connected pins 1 and 2 together and connected them to one side and pin 3 to the other to get a single diode in antiparallel? I agree that the input is largely unaffected for normal operating with the diodes and of course the reduction of RF gain will help with the out of the ordinary plus there is always the option of switching an attenuator although I have so far found this unnecessary with the LNA available even with my extraordinary levels of broadcast signals.

73, Graeme zl2apv

[in3otd]

Hello Graeme,
actually I used two separate diodes also for the MMBD7000 antiparallel configuration, of course just one package with the two diodes connected as you described could be used.

For reference, the diodes connections were like this

73 de Claudio, IN3OTD / DK1CG