Suggestions For a Radar Pulse Amplifier

Hello Peter,

I looked at the previous post in the Spice model forum to get to speed faster.

Since you are intending to use for a +5V supply, the simplest would be to start with a decompensated amplifier such as the LMH6629.  The OPA847 is another option, but you will need to generate a -5V supply as well.

To achieve the best possible noise peformance while limiting the input drive requirement of you low impedance stage, I would recommend using the LMH6629 in a non-inverting gain configuration and have the gain as high as possible.  As John mentioned in the Spice Model forum, the feedback resistor cannot be selected to be too high.  For these high GBWP amplifier I would limit it to 1.5kohm.  The LMH6629 on page 10 shows the frequency response for each package with feeback resistor value varying between 240ohm to 1.5khm.  The peaking is most likely due to the feedback resistor parasitic capacitance interacting with the open-loop gain and the inverting input capacitance.

Since you are looking at a 5MHz BW, achieving ~50MHz -3dB BW would be sufficient, so the maximum gain you can achieve is ~78V/V on the input stage only.  Let's target 50V/V for the first stage to avoid component value to be too small or too large.  In an non-inverting configuration of 50V/V, Rf = 500ohm, Rg = 10.2ohm.  With 50V/V gain, you will achieve 78MHz bandwidth.

Since the first stage will dominate the noise contribution, any subsequent stage can be designed to be capable of providing high gain.  For high gain, high BW configuration, I would recommend using the OPA684.  Each amplifier can be configured at a gain of 100V/V and achieve 70MHz BW. 

The 3 gain stage structure above will then have 500000V/V and 36.8MHz BW.

Your example showed a gain of 100kV/V and your requirement shows 1MV/V, so I settled to 500kV/V.

Now the hard part of the design will be to make sure that it is stable.  As each satge has high gain, there is a possibility of positive feedback through the power supply.  The best approach to minimize the risk is to power the last amplifier in the signal chain first and insert a inductor between the stage (n+1) and the previous stage (n).6644.High Gain Signal Chain.TSC

I attached a possible solution in the TINA file above.  Note that I showed only the inductor in the positive supply, The negative supply voltage if not low enough impedance may require the same inductor.  Also, since the amplifier is used in non-inverting configuration, a mid-point reference buffer such as the BUF602 may be needed as well to avoid any pickup in the GND pin through the gain resistor.

For filtering, since you have multiple stage, inserting an RC filter after each stage may be the most effective and least complicated way of implementing it.  If you want to implement filtering in the OPA684, consider the Sallen-Key as the op-amp is a current feedback amplifier.  Sallen-Key filter will allow you to have gain in the filter stage as well.

Xavier,

Thank you.

Peter,

AM2 was a leftover from another circuit.  It has no use at all.  I apologize for the confusion.

I kept the 100uF and did not see the settling issue you had.  I attached my simulation with the result for reference.

From a nosie perspective, the limit is going dictated by the input stage (LMH6629).  Considering that only the voltage noise contribute to the output (a valid simplification due to the high gain and small resistances used), the callculate output voltage noise density comes out to ~39nV/rtHz at 1MHz.  Simulating it it comes out to be 40nV/rtHz.  Moving onto the SNR simulation, this give ~15dB at 5MHz.

4370.High Gain Signal Chain.TSC

Note that I inserted RC filter after each stage to limit the ferquency response to ~10MHz.  I would also recommend loading each amplifier with 100ohm to GND in practice.

68dB SNR for teh OPA4820 seems reasonnable at unity gain configuration, but in this case, the use of high-pass filter does not make sense to get rid of the noise as the broadband noise dominates.  Due to the low frequency range, low pass filter filter would be best.  Bandpass filter is another possibility but I do not expect the increase complexity of the filter to provide significant improvement over the low pass filter.

Xavier,

Peter,

Thank you.

I have an EVM for this circuit that should arrive within the next few days, so I will be able to verify this on the bench.

You may have an oscillation in the circuit.  Have you looked at the performanc eof the first stage by itself?

The input of the LMH6629 cannot be connected to ground as the bias current will charge the cap.  I would expect a resistor in parallel with the 100nF cap, say 100ohm.

Best regarsd,

Xavier

Pete,

The circuit works.  We achieve ~100dB of gain.  See attached results.

In order to evaluate it on the bench, we had to place a bandpass filter at the frequency of interest on the output of the arbitrary waveform generator, followed by two 40dB attenuator to get the signal low enough.

To isolate the sinewave, we averaged it 512x.  See results for 1MHz and 5MHz.  Another 1MHz result shows the noise in the signal.

2548.Vout_Sig3.xlsx

I also included the layout we used for the evaluation.  Let me know if you want me to ship you the board used.

8562.High Gain Layout.pdf.

Best regards,

Xavier

Xavier,

I think I am getting similar results. 

Would it be possible for you to collect samples of the output with no input signal?  I am measuring about +/- 1.25 volts or about 2.5 volts P-P with no input.  To me that is all noise.  Oh, there are no oscillations.  The signal is just random noise.  The image below was collected by triggering on the noise and raising the scope trigger level until it just barely triggers.  It is then triggering on the highest peaks, ~1.25 volts.  So we see just noise unless the noise pulse is higher than 1.5 volts or so.   Note there is no repeating of this signal like it would if there was oscillations.  Pete

I forgot to mention two tiny items.  On your spreadsheet, I believe the horizontal axis for the top and bottom graph should be us not +Vin(mV).  Like I said, tiny and minor.

Pete

Pete,

Yes I believe that this is noise.  I eliminated the 3rd gain stage (set it to 1V/V) and now only have the first gain stage and second gain stage set at 100V/V for the LMH6629 and 50V/V for the OPA684 for a total of 5000V/V.  The signal below does not exceed ±50mV.

7750.No Input signal (100V-V LMH6629 + 50V-V OPA684.xlsx

I will now be looking into noise amplified added by the last stage.

Xavier

Xavier,

I guess the question comes down to, "Is the measured noise similar to what TINA predicts or is something else happening?".

I added some things to your excel sheet:

I multiplied the readings by 40 to get back the gain of the last stage,

Counted how many samples exceeded +/- 1 volt.  This is an arbitary value but larger than I can accomidate.

2,356 samples out of 10,000 were > 1 volt or 24% of the time.

Does that agree with TINA S/N?

3821.NoInput 200000V-V.xls

Pete

Pete,

Looking at the Total noise (RMS output noise) in TINA, I get 15.9mVrms at the output or 2.899uVrms input referred (5482.77V/V).

Evaluating the same output RMS noise with the socilloscope I get 16.1mVrms at the output or 2.876uVrms (5597.36V/V implementation).

So there is a strong correleation between noise between TINA and the implementation.

Best regards,  Note that this was measured for 0ohm source.

Xavier

Xavier

I'm missing where the 5483 V/V  or 5597 V/V comes from. 

This is the circuit 4370.HighGainSignalChain[1] right?

48.9 V/V * 49 V.V * 39 V/V =93676>>>>99db

Pete

Pete,

To avoid clipping on the output and eliminate complexity in the measurement, I set the last stage at unity gain.  The first stage was set at a gain of ~100V/V and the second stage at ~50V/V.

The actual gain simulated was 5483V/V and the gain of the board was 5597V/V.

Pete,

The high speed comparator would certainly help reducing noise susceptibility.  Another method to improve nosie would be to use a transformer at the input, so you can add some noiseless gain.  I was think of using a 1:4 turn ratio (1:16 ohm ratio) as soon as I can get one in.  I was looking at the ADT16-6T from Minicircuit.