Filtering and amplifying a 5mV signal

How noisy is your signal, and how much of that is above 8MHz? You have plenty of space, so can afford to use lots of parts.

Sallen-Key / VCVS filters are simple, one op-amp per two poles. We talk about them on pages 407-410, and show parameter tables for major filter types. These use non-inverting amplifiers, with specific gains of from G = 1.084 to about G = 2.904.

But working on a crowded PCB, I reasoned we could use a simple version with higher gains, by reducing the feedback capacitor. Aha, this worked out very well for G=10; it's easy to make a one op-amp 3-pole low-pass filter this way. It may not have had mathematically-accurate specific-filter properties, but it was pretty close, with good bandpass, good rolloff, etc.

An opamp in a G = 10 filter with say 15MHz of bandwidth, and decent signal-handling capability (high slew rate), would need to be a CFB type. Their required low-value feedback-resistor requirements fit well with the non-inverting configuration.

Using this scheme, you could accomplish your goals with just one opamp. If you need G=100, use two opamps, or a dual IC.

At 8 mv, expect ground loop problems. 1 LSB will be a couple of microvolts. Expect many LSBs of noise.

Is that 8 MHz signal narrowband? What's the source impedance? A tuned network could buy a lot of s/n.

You might start with a good diffamp and then filter at higher level.

This board

has a 250 Mhz 12-bit ADC. The input is differential, and after the diffamp is a 5-pole LC lowpass filter, then an official ADC driver chip, then the ADC.

have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 things, filter it, and amplify it, the end result needs to fit on a fairly small p cb... something like 2" x 2" , I'm using FilterPro right now to come up wit h some filter ideas... do you think it's ok to do an active filer that can do the gain and bandpass? Then feed that into an ADC... I'm looking at usin g an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly appreciate d!

First, is the noise you want to filter larger than the signal? If not you might be able to amplify first and filter at the end. This will require am plifier bandwidth enough to prevent the noise from causing distortion.

If the noise is larger you have to limit the initial amplification so the n oise does not over drive the amplifier. Then filter and amplify some more to reach the signal level you need.

I don't know that 5 mV is so low you need to take special precautions to li mit intrusion of noise, but if so guard rings will help isolate intrusion f rom other signals on the board. This largely depends on the details of you r design.

You don't indicate the source of the signal. If it is high impedance you w ill need to take special precautions to avoid letting noise into the signal in the cable to the board. A good shield is essential and that shield may need to be a guard band as well.

One final note, for an active filter to work well the gain bandwidth produc t of your amp needs to be much higher than the frequencies you wish to filt er. Once the open loop bandwidth of the amp is reached the filter pretty m uch stops working.

With an 8 MHz signal it might be better to use a passive anti-alias filter, over sample the signal and then perform further filtering in the digital d omain. Not sure how fast you want to sample or what will receive your data , so this may not be practical. You mention a 65 MSPS ADC, but that doesn' t mean you intend to sample at 65 MSPS.

I don't know if it is practical with an 8 MHz signal, but delta-sigma ADCs provide their own oversampling which allows simpler anti-alias filtering. I haven't worked much with them in a while and don't know where they top ou t these days.

What does the wanted signal look like ? Is it 0-8 MHz sine, a complex waveform 8 MHz signal (lots of harmonics) or a narrow band signal around 8 MHz ?

In the first case a pi-filter /CLC) low pass filter at about 10 MHz should be enough if 65 MHz sampling is used to suppress aliases. If a lower sampling rate is used, more filter sections may be needed.

In the 8 MHz fundamental case and 65 MHz sampling a multi-section pi-filter at 25MHz would be needed.

If 8 MHz narrow band signal is used, a bandpass LC filter could be used. a 10.7 MHz IF transformer loaded down to 8 MHz with external capacitors on both input and output LC side should be enough. The advantage with bandpass is that undersampling can be used, so maybe 1 MHz sample rate might be sufficient.

The 5 mV signal is quite strong (at least for the bandpass case) . so the amplifier could be after the lossy filter, thus avoiding the risk of amplifier overloading. If two stages are needed to get sufficient gain to run the ADC into full scale, put one low gain stager before the filter and a higher gain stage after the filter, so that the filter sees a well known and constant impedance on both the input and output.

have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 things, filter it, and amplify it, the end result needs to fit on a fairly small p cb... something like 2" x 2" , I'm using FilterPro right now to come up wit h some filter ideas... do you think it's ok to do an active filer that can do the gain and bandpass? Then feed that into an ADC... I'm looking at usin g an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly appreciate d!

Knowing the bandwidth would help.

have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 things, filter it, and amplify it, the end result needs to fit on a fairly small p cb... something like 2" x 2" , I'm using FilterPro right now to come up wit h some filter ideas... do you think it's ok to do an active filer that can do the gain and bandpass? Then feed that into an ADC... I'm looking at usin g an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly appreciate d!

Thanks for all the replies, my signal is very sine wave-ish, it's getting l ost in noise that is around the same frequency of interest, it's coming off a coax cable, 50ohm source

I've pulled up the VCVS section in my AoE book, seems pretty do-able, it wo uld be great if I could get a couple stages with G=10

Does using a real fast op-amp increase my chances of getting more high freq uency noise? I've never used a CFB amp, I'll check them out

Noise is kinda the main issue we're dealing with, should I consider using a diff-amp to bring in my signal (similar to the order Larkin described) dif f-amp then filter then ADC driver?

I don't understand undersampling enough to mess with that

Much Thanks!

I have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 thing s, filter it, and amplify it, the end result needs to fit on a fairly small pcb... something like 2" x 2" , I'm using FilterPro right now to come up w ith some filter ideas... do you think it's ok to do an active filer that ca n do the gain and bandpass? Then feed that into an ADC... I'm looking at us ing an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly apprecia ted!

lost in noise that is around the same frequency of interest, it's coming o ff a coax cable, 50ohm source

would be great if I could get a couple stages with G=10

equency noise? I've never used a CFB amp, I'll check them out

a diff-amp to bring in my signal (similar to the order Larkin described) d iff-amp then filter then ADC driver?

Filters remove noise signals from signals of interest based on a difference in frequency. If they are about the same frequency it will be hard to sep arate them with filters. Looks like it will be more useful to prevent the noise from getting into the signal if possible.

To maybe make the issues more clear, I would refer to both the noise and yo ur signal of interest as signals since they will be treated the same by the filter, the coax, the ADC, etc. So signal of interest and noise signal.

Do you know the source of the noise signal? Is it entering the same way yo ur signal of interest is at the transducer? Or is the noise signal enterin g through the coax or a grounding issue?

You can think of undersampling to work a bit like mixing. The frequencies of any signals in your input will be mixed into the band from 0 to fs/2. E very other band of fs/2 width will be inverted in frequency, not that it ma tters in your app I think. But if your signal of interest is at fs/2 plus a small increment and the noise signal is at fs/2 plus a larger increment, you will separate them in the digital domain by a high pass filter.

I think it may be much easier to prevent the noise getting into the system if you can.

Table 6.2 in AoE III presents math-derived parameters for standard Sallen-Key filters, such as a Butterworth filter, with inconvenient K values (non-inverting op-amp gains), from 1.1 to 2.3,etc. But I've found you can set K=G=10, if you dramatically increase C2 and decrease C1. Here's a schematic plus four rolloff plots, with formulas.

Page 410 says cutoff freq, f = 1 / 2pi R2 C2 sqrt(mn), but the plots show low C1/C2 ratios seem to overstate f. I'd say keep m=1 (R1=R2), and n = C1/C2 = 0.16 values, and scale R*C values to place the cutoff where you want.

One issue with this trick, the filter is very sensitive to your exact C1/C2 ratio, quickly going from low Q with sluggish rolloff, to high Q, with peaking. See plots as C1 is changed from 120pF to 180pF. But you do get G=10, and in your application the rolloff shape may not matter.

The trick is to use all the signal power that's available and jam it into an amplifier. If the signal is 50 ohms and reasonably narrowband, impedance-match it into a low-noise semiconductor, a bipolar or phemt or even an opamp. That needs an LC tuned network first thing. You could easily get a noiseless gain of 5, maybe 10, from a tuned network.

A mediocre FM or shortwave receiver can play nice music with microvolts of signal.

If the signal comes in on a coax, a diffamp won't help.

The OP's signal is at 8 MHz. Can you use make an opamp active filter up there? A gain of 10 and a Q of ten seems like too much. Even for just a band pass Q of 10 at 8 MHz, what opamp would you use?

George H.

For a low-pass filter, you probably want quite low Q, near 1.

Yes, G = 10 at 8MHz implies an op-amp f_T well over 80. You might well want to separate the gain and filter functions. But the non-inverting gain nature of Sallen-Key VCVS filter stages lends itself well to using CFB opamps, where G=10 at 8 MHz is an easy spec, and it's fine to also be a filter.

Your choice would depend on things like available power, space, $$ budgets, voltage range, etc. Tables 5.4 and 8.3c list some CFB types. A new CFB table in the x-Chapter book has many dozens more.

Looking through table 5.4 (pg 310) and table 8.3c (pg 524)

I have 5V available, $ budget is available for higher end parts, it's running off a battery so lower power is better, unless trade off with noise is too much

looking through the tables these seem plenty fast, I guess I should focus on power and noise parameters... maybe the AD8000 or LMH6723

What's the x-chapter book?

Thanks!

You still have not specified what your wanted signal looks like. You have not specified what the noise is like. Without addressing these issues, it is impossible to give any meaningful suggestions.

You claim that the signal is sine wave-ish so this would suggest that it is a signal around 8 MHz and a suitable _band_pass_, filter centered at 8 MHz would be the answer. But what is the spectrum of the wanted signal, i.e. how much is it spread around 8 MHz, is it 7.999 MHz to 8.001 MHz or 7 to 9 MHz ?

What is the noise like, is it just thermal (white) noise or are there some discrete interference frequencies close to 8 MHz. If just a single interference frequency, a _band_stop_ filter might be adequate to filter out the interference.

Anyway, the 8 MHz is sufficiently high frequency to implement passive LC band pass or band stop filters within the available space.

Without specifying these issues, you will not get any meaningful answers and it is far too early to think about any specific ADCs.

Sure you could sample at 25 MHz and using DSP to filter out the wanted signal, but the power dissipation would be high.

Problem is, once you get noise into your signal's bandwidth, no amount of processing will take it out. A quiet front-end is the best start.

As you note, the signal nature and bandwidth need to be specified.

n. I have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 thi ngs, filter it, and amplify it, the end result needs to fit on a fairly sma ll pcb... something like 2" x 2" , I'm using FilterPro right now to come up with some filter ideas... do you think it's ok to do an active filer that can do the gain and bandpass? Then feed that into an ADC... I'm looking at using an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly apprec iated!

ng lost in noise that is around the same frequency of interest, it's coming off a coax cable, 50ohm source

Something tells me the OP is not willing to specify them here. In which cas e one can only state the obvious: any filter can only discriminate between signal & noise in ways that they differ, whether that's by frequency, ampli tude, polarity, pulse shape, bit pattern, repetition, synchronousness etc. You the OP presumably know in what ways they differ and therefore in what w ays they could possibly be separated. AFAIK we can only help once you tell us what the differences are between signal & noise.

NT

The 'frequency of interest', does it have some slow modulation that you care about? There's tricks that might help, like phase-locking to the signal (and using the phase-locked reference to selectively amplify the tracked sine, with much less amplification of the noise). If it has fast modulation, though, your 'signal' has enough bandwidth to be harder to separate from any bothersome background.

After phase-locking, you can mix down to nearly DC, and use filters that are easier to build or buy (intermediate frequency filtering, a common bit of RF technology, can be done digitally nowadays, too).

n. I have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 thi ngs, filter it, and amplify it, the end result needs to fit on a fairly sma ll pcb... something like 2" x 2" , I'm using FilterPro right now to come up with some filter ideas... do you think it's ok to do an active filer that can do the gain and bandpass? Then feed that into an ADC... I'm looking at using an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly apprec iated!

ng lost in noise that is around the same frequency of interest, it's coming off a coax cable, 50ohm source

If it's not white noise, but spikey interference/pickup, in principle you can filter it. But I agree best to keep the front end as quiet as possible .

If you 'know' the source frequency you can do a lockin type thing... a mixer at 8 MHz I assume.

Sorry, just thinking of 'other cases', not knowing what the OP is about. George H.

. I have a 5mV 8MHz signal I need to feed into an ADC, I want to do 2 thin gs, filter it, and amplify it, the end result needs to fit on a fairly smal l pcb... something like 2" x 2" , I'm using FilterPro right now to come up with some filter ideas... do you think it's ok to do an active filer that c an do the gain and bandpass? Then feed that into an ADC... I'm looking at u sing an AD9238 (12-bit, 65MSPS) ADC... any tips/thoughts is greatly appreci ated!

g lost in noise that is around the same frequency of interest, it's coming off a coax cable, 50ohm source

would be great if I could get a couple stages with G=10

requency noise? I've never used a CFB amp, I'll check them out

g a diff-amp to bring in my signal (similar to the order Larkin described) diff-amp then filter then ADC driver?

What's a tuned network with gain? I've never done an LC bandpass filter bef ore, but put one together and ran it in LTSpice, and it looks pretty good, would just an LC bandpass filter work for this kind of thing?

What parameters do I need to watch out for on my inductors when doing an LC filter like this?

My LTSpice Filter:

I don't have a lot of info about the issue, a friend was having an issue re ading back data from a sensor and was having to average a bunch of samples to see his signal, seemed like an interesting problem so I've decided to ta ke a crack at it, and learn something along the way

Much thanks!

That filter is OK as such, but it loses half of the signal voltage

It's simple: 50 ohm coax in, series capacitor, inductor to ground, with a hi-z amplifier at the LC junction. The signal voltage is amplified by the Q of the LC circuit. Q is limited by your signal bandwidth requirement and eventually by the inherent Q of the inductor. Q=10 or maybe 20 is reasonable if your signal is narrowband, which gives a free, noiseless voltage gain of 10 or 20. The LC resonant frequency should be tuned to peak the resonance at 8 MHz.

If you need a flat, wider passband, a more complex network can do a real bandpass function but still step up the impedance and get some voltage gain. Your waveforms don't look narrowband, so a simple peaked LC may not be ideal. We don't know anything about the signal.

The amp could be a jfet or a phemt or a fast jfet opamp, something with just a few pF of capacitance.

If you connect the 50 ohm coax directly into a hi-z amp, even through a bandpass filter, very little of the available signal power is pushed into the amp, which is a waste.

Alternately, Mini-circuits makes some 50-ohm matched/untuned/wideband mmic amps that have low noise figures, below 3 dB at 8 MHz. That's a noise density below 1 nV/rootHz, which would be