Magnitude value in AC simulation with Cadence Virtuoso

Dear Senan,

Senan said:

I usually set it to "1" so I can easily interpolate the output signal as the gain of my amplifier as simply the output is equal to gain when input is one.

This is correct.

Senan said:

However, it just came to my mind if this value has an effect on the amplifier small signal characteristics or not, just I want to verify this point.

In an AC simulation, the devices are all modeled as linear devices with small signal parameters derived from the initial DC operating point of each device. As such, the resulting response will not show any non-linear effects with input signal amplitude - but will only linearly scale the response you are studying by the relative magnitude of the input stimuli. I hope this makes sense Senan.

Shawn

Dear Shawn,

Thank you for your help, I fully understood your answer, 

Since I am now characterizing our amplifier practically, I am performing the AC response of the closed loop gain under different gain setting,

This task I was performing it in Cadence with the AC simulation as the way I mentioned in my first post.

If I want to make comparable practical measurement, do I have to use a test signal with a small amplitude? 

I found however, even with small signal test of 100 mV and large gain setting, the output signal will be large and in my opinion it will not be as identicall to Cadence setup, may be I am wrong

Thank you

Senan said:

If I want to make comparable practical measurement, do I have to use a test signal with a small amplitude? 

I found however, even with small signal test of 100 mV and large gain setting, the output signal will be large and in my opinion it will not be as identicall to Cadence setup, may be I am wrong

Since you are simulating a small-signal linearisation of the circuit with an AC analysis, the size of the AC input signal doesn't matter, because the response will be linear. If you're comparing that with a real-life measurement, the circuit won't respond in a linear fashion if the input signal is too bug - because it will saturate in some way and clip. You can absolutely simulate with a small signal value which meets the actual circuit's limits for responding linearly, and you should get similar answers. I'm not sure why you think otherwise?

Andrew

Dear Senan,

Senan said:

If I want to make comparable practical measurement, do I have to use a test signal with a small amplitude? 

I found however, even with small signal test of 100 mV and large gain setting, the output signal will be large and in my opinion it will not be as identicall to Cadence setup,

As Andrew correctly stated, if your simulation model accurately captures your circuit, a lab measurement of its transfer function and gain versus frequency with a "small-enough" input signal will provide the same transfer function and gain as you observe in an AC small-signal simulation.

However, let me add a few caveats that may lead to unexpected measurement results if I may for your consideration. These may or may not be relevant, but have been factors in measurements I have performed.

1. The AC simulation does not include DC with the exception of the use of its DC operating point to determine small-signal device parameters. However, in your lab measurement, there will be some finite DC offset due to mismatch of devices. You can think of this as an ideal DC source of value Voffset in series with your input signal. Therefore, if your amplifier DC gain is Av, this DC offset will cause an output voltage of:

Vout = Av*Voffset  

and will be limited by your output voltage limits of Vmax and Vmin. For example, if Voffset = +10 mV and Av = 100 and the absolute value of Vmax and Vmin < 1, this will produce a DC output of +1 V. Clearly, this cause your measured gain to differ from your simulated gain as the output devices in your simulated gain do not include this 1 V offset from their nominal 0V input level. Hence, when you perform your gain measurement, you must try to minimize the impact of this DC offset by trying to zero it out, or make sure it is small enough not to impact the DC output voltage from its desired value with an input of 0 V.

2.  From your description, the gain of your amplifier is variable with some type of setting. I might suggest you start your measurements with the gain set to its minimum value. Perform measurements at several gain settings starting with its minimum value. For each gain setting, record the DC output voltage to validate it is close to its quiescent value.

3. Your 100 mV input, to me, is considered a "large signal". In small-signal measurements I have performed on amplifiers with any type of gain I choose input amplitudes on the order of 1 mV to 10 mV - or less sometimes. If your signal generator does not provide this signal level, I have often used an attenuator to drop the level to the value I wish to achieve.

4. Finally, when performing a gain measurement, I always measure both the input and output with the same piece of test equipment. In essence, a gain measurement when you "assume" the input level is at a specific level does not include any measurement error of your measurement system and may not reflect any loading of your amplifier on the signal generator.

I hope this helps provide a few things to consider Senan if you have not already!

Shawn

Dear Sahwn, Dear Andrew,

I am grateful to you both for this amount of help and practical assistance beside your Cadence support,

I agree on all the point you mentioned, indeed my main problem you both have came to it, that is the  definition of "small-enough" input signal because by any way I am not going to test with a signal that clip my amplifier, that is absolutely prohibited.

Also as you mentioned, the offset voltage is well trimmed/calibrated before doing the test and we actually measure input/output with the same piece of measurement.

Coming to the main point again  definition of "small-enough" input signal. I am working with 3.3 V technology, consider that offset is nulled to zero and my gain setting is varying between unity up to maximum gain of 128 (managed in binary weighted steps). 

I am applying 100 mV for small gain, and when I reach the gain of 128 I am applying 10 mV. In either case the output signal is not saturating but when the gain is unity the output is 100 mV and when 128 is 1.28 V , which is not saturating but  NOT small.

After your answer it comes to my mind another criteria for defining the meaning of small signal, that is the output signal should not experience slew rate limitation, specially when sweeping the frequency at the higher region of the Bode Plot, I think it is true since you also mentioned that AC simulation with Cadence consider the linear model based on the DC operating point.

So the less possible input signal, the better output signal that avoid any source of non-linear behavioral (clipping/slewing).

But Shawn testing with 1 mV is difficult not because of the source signal generation, but due to the fact it becomes near to the noise coupled to my inputs/outputs traces with my PCB setup.

Thank you Andrew and Shawn very much once again

Best Regards    

Senan said:

I am applying 100 mV for small gain, and when I reach the gain of 128 I am applying 10 mV. In either case the output signal is not saturating but when the gain is unity the output is 100 mV and when 128 is 1.28 V , which is not saturating but  NOT small.

From a simulation perspective, the magnitude of the input or output signal is irrelevant; you could apply a 1000V input if you wished, and the output would be 128,000V; the simulator is simulating a linear model so it cannot clip.

The fact that with an input signal of 10mV in real life you get an output voltage of 1.28V is also (potentially) OK. All the "small-signal" from analysis perspective means is that the simulator will give a response assuming that the circuit is perfectly linear under that bias condition. It doesn't have to be "small" as such, and certainly not "small" at all nodes in the circuit - the point is that you are observing a linear response.

What value you pick (for measuring on the bench) will be entirely dependent upon your knowledge of how the circuit is supposed to behave with the signal magnitude you've given; presumably you will be aware of large-signal effects by having simulated in a transient analysis (or even a pss/hb analysis) to observe if there is any distortion at the signal magnitude;

Andrew

Dear Senan,

Thank you for the update and information about the specifics of both your measurement technique and observations! The information was quite useful!

Senan said:

So the less possible input signal, the better output signal that avoid any source of non-linear behavioral (clipping/slewing).

But Shawn testing with 1 mV is difficult not because of the source signal generation, but due to the fact it becomes near to the noise coupled to my inputs/outputs traces with my PCB setup.

You are correct on all counts Senan! Any slew-rate limiting that your output stage is an inherently non-linear process is not modeled by an AC simulation. It will be included in a transient simulation as Andrew noted. Hence, as you stated, it is best to avoid any combination of input amplitude, frequency, load, and gain setting that induces non-linear slewing of your output

Please allow me to add another possible means of measurement which came to me after I responded yesterday that may help. You did not note what specific piece of test equipment you were using for your measurement(s). From your description, I am thinking it may be some type of DSO or oscilloscope.

When measuring small signals with a DSO or oscilloscope, your sensitivity (which limits measurement accuracy) is determined by the noise. Since you may be using probes, the noise is the integrated noise over the bandwidth of the probe - which can be quite large.

The standard practice when the need to measure a transfer function accurately (< 0.01 dB), as an alternative, you might consider using a spectrum analyzer in lieu of your oscilloscope. You can adjust the resolution bandwidth of the measurement to reduce the measurement noise bandwidth significantly. Further, you will quickly be able to tell if your output signal is exhibiting non-linear behavior as harmonic distortion spurs will appear in the spectrum. It is also likely the crosstalk due to adjacent traces will not be at the same frequency as your injected input frequency and hence will not impact your measurement accuracy.

If you do choose to try this, when making gain measurements to compile a transfer function, make sure you use the same resolution bandwidth settings for your input signal measurement and output signal measurement to eliminate any difference in the measurement bandwidths of the two measurements. You may not be able to use a 1 mV input signal, but you should be able to use a smaller signal than with an oscilloscope measurement since the noise bandwidth will be far less.

Shawn

Dear Andrew,

Dear Shawn,

Thank you very much for your kind explanation, you made the principle of AC simulation crystal clear for me and I have no further question on it. 

For the practical measurement, I am using DSO unit from Rohde&Schwarz RTM3004 with 10 bit resolution, Bandwidth of 500 MHz and automatec sampling frequency of 5 GS/s. It is actually mult-purpose DSO with logic analyzer capability and can perform frequency spectrum and bode plot. I am setting both channel interface to 1 M Ohm because my amplifier can not drive 50 Ohm.

Unfortunately we have an old spectrum analyzer and I am not in favor of using it :)

One interesting thing I may share it with you, my circuit is fully-differential amplifier which generally work perfect when we apply a fully-differential input and characterize the fully-differential output signal. however, the signal generator of our DSO is single-ended (and I think there is no DSO with differential source option). So what I am doing is that I am connecting one of the differential amplifier inputs to the VCM and reading from one of the two outputs, in other words I am treating the fully-differential amplifier as a single-ended amplifier. Certainly I am expecting to loose half of the selected differential closed loop setting and that is not a problem, unless if it has effecting the amplifier bandwidth performance which I am investigating for the moment.

I didn't use balun transformer because it will limit the DC response of my amplifier and I am not aware of available balun that can go near to DC region.

Thank you once again and I am happy to share with you the experimental work

Best Regards  

Dear Senan,

Once again, I thank you for taking your valuable time to include the added measurement information.

Senan said:

Unfortunately we have an old spectrum analyzer and I am not in favor of using it :)

OK. I hope you are not avoiding it just because of its age! Speaking for myself, I've more than enough years for me and I like to think I can still be useful at some things!

Senan said:

I didn't use balun transformer because it will limit the DC response of my amplifier and I am not aware of available balun that can go near to DC region.

for your back pocket if you ever need it, Figure 1 is what I use to convert a single-ended signal generator output to a differential signal with his signal integrity. The bias-tees are used to add in the common mode. Unfortunately, Picosecond Pulse Labs was purchased by Tektronix and they are no longer selling their passives. However, there are till some available for purchase on-line. Of course, other vendors also provide them, but in my experience, I found their products exceptional.

Shawn

Figure 1

Dear Shawn,

Thank you very much for your feedback and your suggested balun, it is indeed masterpiece to see a balun with such a range, I could have find some people selling it in ebay and I requested them to buy it, looking very interesting.

I do appreciate your kind support and help

Have a nice time

Best Regards