128Hz Bandpass Filter 2

[JBirch]

Need to build / buy a 128Hz band pass filter. Having trouble finding any information regarding design, components required etc.

Needs to have a gain of 1, and a bandwidth of 10Hz. Needs to be an active filter to keep losses to a minimum.

Thanks heaps in advance,
Jamieson

 

[Skogsgurra]

Is it a "signal" bandpass or a "power" bandpass? You mention power loss and that makes me wonder. I think it would be better if you describe the application; what is the source impedance, what is the load impedance? What power levels and what voltages levels are involved?

Gunnar Englund

http://www.gke.org

 

[JBirch]

The intended application is measuring of step and touch potentials associated with power system earth faults.

We are injecting a 5A 128Hz signal into a remote earth stake and measuring the potential rise. We need to filter out any power system frequency voltages and associated harmonics. ie 50Hz, 100Hz, etc

This is why we use 128Hz as it is removed from any of the above frequencies.

The input voltage could be anywhere upto 10V in bad soil conditions. However, usual volatge inputs are around the 100mV range.

Load impedence is input to Digital Multimeter. Source impedance will vary with soil conditions, etc

Cheers!

 

[Skogsgurra]

There are lots of "cook-book" recipies for active filters that do what you want.

A very efficient filter topology is the opamp with a Twin-Tee in its feed-back.
Parallel the Twin-Tee with a resistor to get a well defined gain. Remember that band-width is the -3 dB point and that there may be enough unwanted signal left at 100 and 150 Hz to influence the measurements.

With the source (1 - 100 ohms, probably) and load impedances given, I think that a passive LC filter would work quite well. It has advantages like no battery needed, no saturation (within limits) and usually more robust.

If you want to have full control, go DSP or use one of the ready-made filters offered by several IC manufacturers.


Gunnar Englund

http://www.gke.org

 

[ScottyUK]

If I'm interpreting this correctly, the filter is only in the received signal path, not the power circuit. You best choice for high suppression of the stop bands would be a digital FIR or IIR filter. These are available as building blocks with a high level of integration and small external component count.

The switched capacitor filters are also quite effective at low frequencies.

A gyrator based filter is also a possibility. Gyrators are effectively synthesised inductors using op-amps and capacitors to produce 'inductors' with very high Q-factors which could not be built as wound components (except maybe using superconductors). These allow filters with very high Q and thus very steep cutoff to be constructed.

----------------------------------

I don't suffer from insanity. I enjoy it...

 

[jimkirk]

Another question to ask for your design is how much side band rejection do you need?

How accurate are your frequencies? If the center frequency of your filter is a few hertz off from your signal, you might drop a few dB. If that's an issue, you need to design a filter with a flat pass band.

Check out

http://www.linear.com/pc/categoryProducts.do

Linear makes a variety of switched capacitor filter building blocks that should do the job.

Jim

 

[VEBill]

There are other approaches than just plain old band pass filtering around 128 Hz and hoping for the best. The simple filter architecture that you've selected is, in radio-speak, just an old TRF (Tuned Radio Frequency) receiving technique from the 1920s. There's been a lot of progress since then.

You could simply A/D the whole thing at the receiving end and then use an FFT to ferret out your 128 Hz signal. It doesn't seem like you'd need too many bits for the sorts of signal levels you've mentioned: 16-bits 96dB would seem to be plenty. This approach might be extremely applicable if you're planning to use a laptop anyway to log the data. The solution becomes mostly SW with an off-the-shelf A/D card and a cable.

You could (should?) also modulate a code onto your 128 Hz probe signal to help the receiver pull it out of the noise. One example that springs to mind is a Barker Code. Using codes would allow your receiver to dig the signal out of the noise floor. There are coding techniques that allow endless integration. Modulating a code is an extremely powerful technique. The only downsides are maybe extra cost and the time to acquire. Might take several seconds to acquire the code when the carrier is only 128 Hz.

I could go on & on... In summary, if you look at this problem from the view of radio (think: 128 Hz radio link over a wired path), there are a brazillion techniques to extend the capabilities beyond what could be done using a plain old 'TRF' filter.

You should assume that there will be some unknown noise floor even at 128 Hz. Do NOT assume that the frequency is silent other than your signal. Sometimes the noise will be very high amplitude. Make sure your system can deal with the worst case, even if just an error message about noise.

If you don't use an advanced code with this capability built-in, then you should probably include an automatic system to help distinguish your signal from the noise floor in the same bandwidth. Maybe an on-off keying cycle. If you don't, then there would have to be two operators with walkie-talkies yelling back-and-forth, "Turn it off for a second. Is it off? Oh darn, it's still there. Are you sure that it is off?" etc.

Then again, simple filtering might do the job most of the time.

 

[2dye4]

I agree with above regarding rejection of signals
in the 128 hz band that not created by your test.

You could build your source to cycle on/off at a 1 hz
rate. Set up the FFT to sum the energy while your signal
is on and subract the energy when your signal is off.
That way you can reject energy from other sources.

Many simply pc data aquisition devices come with FFT
analysis programs.

 

[logbook]

I think you requirement is very asymmetric with respect to cutoff. You need lots of attenuation at lower frequencies and hardly any at higher frequencies. I would therefore suggest a many pole high pass filter and perhaps a single pole roll off above say 300Hz. This will be much easier than a symmetric bandpass filter.

Download a free copy of FilterPro (Texas Instruments) and play with it. You wil see that a Butterworth 6 pole filter still only has 10dB attenuation at 100Hz if you set the 3dB corner at 120Hz. You need to establish some facts about your system. How much 50Hz do you need to filter out? "All of it" is not an answer. Will 45dB do. If not then you need more than 6 poles.

 

[ScottyUK]

I would think a Chebychev design would be better suited to this application - the superior attenuation characteristics from a filter of a given order would outweigh the presence of passband ripple.

I like Logbook's idea about the assymetric filter.


----------------------------------

I don't suffer from insanity. I enjoy it...

 

[Warpspeed]

What worries me about this type of application is that the incoming signal may have a LOT of out of band noise, and this could conceivably be of much higher amplitude than the wanted signal.

Any type of active filter, switched capacitance filter, or the DSP approach may create more problems than it solves. Very easy for the filter to be overloaded, or suffer evil effects from clipping, intermodulation, aliasing and so on. Unless you plan to monitor the signal with an oscilloscope, I would be very wary of anything fancy, even if it has some sort of auto ranging on the input.

My vote goes to a completely passive filter. High insertion loss can easily be made up by some additional gain at the output to bring the whole thing up to exactly unity at 128Hz. It would be immune to overload from powerful out of band signals.

What goes in at 128Hz comes out unmolested.

 

[IRstuff]

One option may be a phase-locked loop, which is very good at rejecting out-of-band noise.

TTFN

 

[ScottyUK]

The relatively close proximity of the likely interfering signals - 50/60Hz and their harmonics which will, or should, be predominantly of odd order, makes it a tough spec for a passive filter to achieve. There's a circuit diagram of a passive high order notch (or bandpass?) filter illustrated in 'Art of Electronics' by Horowitz & Hill. It is fairly complex,for a purely passive circuit and there are a lot of expensive tight-tolerance components with peculiar values in that design. We used this circuit as the basis of a college exercise on the effect of tolerances on circuit behaviour. The 'perfect' circuit was very good when simulated, but the response quickly went to crap as the tolerances opened up. Passive = pricey for this kind of application!

----------------------------------

I don't suffer from insanity. I enjoy it...

 

[VEBill]

WS: "...out of band noise, and this could conceivably be of much higher amplitude than the wanted signal."

Higher, but not really that much higher that a well designed system should be fooled.

OP had stated: "...input voltage could be anywhere up to 10V in bad soil conditions. However, usual voltage inputs are around the 100mV range."

Those are very good healthy signal amplitudes. One could easily pull those out of the noise, even the FULL AC voltage of xxx volts riding the same wire (!), using a normal 16-bit A/D. Nothing difficult about the range or ratio of 100mV to xxx volts.

But he still needs a method to distinguish the signal from the noise floor. A simple filter simply won't distinguish 128 Hz signal from 128 Hz noise. Although 128 Hz may seem (mathematically) like it might be a quiet frequency, this ignores broadband noise. AC power wiring can be a VERY noisy place.

If the system doesn't distinguish (automatically), then the operator(s) will have to do so (manually).

 

[logbook]

I can't see there being any 128Hz noise.

An all passive filter at 120Hz is going to use components that are far too large for comfort. The Sallen-Key filter should do nicely. I have simulated the response on SPICE with a 6 pole Chebyshev high pass filter and a single pole low pass on the output (1K-470nF). It takes 40ms for the 128Hz signal to ramp up to the required steady state level (6 cycles of the waveform). This is the sort of AC transient I was expecting, but which might not be obvious at first glance.

 

[cbarn24050]

WHat you need is a wave analyser. Marconi TF2330, they come up on ebay every so often, go for pennies as not many people know what they are. Or you could just get the manual and make your own, would be easy as you are only using one frequency.

 

[VEBill]

Logbook: "I can't see there being any 128Hz noise."

I'll bet that almost every noise source on the typical power line (or associated running-parallel ground wire) would NOT be characterized as 'narrow band'. Think about hash, static, contacts opening, switching power supplies, leaky insulators, motors, various equipment, etc. etc. etc.

There will be a broadband noise floor, even at 128 Hz or 130 Hz, or 147.7 Hz, or any other number you care to mention. You're not likely to find a reliable 'clear channel' anywhere on that band (radio-speak).

If everyone is thinking in terms of narrowband noises on the power and ground lines, then they'll be disappointed.

The noise spectrum will obviously vary from place to place, so even making the effort to test a few locations would be useless in so far as trying to prove a negative.

Therefore, the system should include a method to distinguish in-band noise from the test signal. Even 'God-like' <insert your favorite filter here> filtering alone will not do that.

System needs on-off keying at the very least. Can be done manually (just need a paragraph in the instruction manual).

 

[ScottyUK]

cbarn might have a point - we picked up an old but perfectly good HP3561A Dynamic Signal Analyser for about £1500. We use it for setting up the voice-frequency tone signalling equipment which runs between our HV substation and generator control package. It's a handy instrument to have around.


----------------------------------

I don't suffer from insanity. I enjoy it...

 

[logbook]

VE1BLL, I take your point. I guess one would need to knock a simple system up and see how bad the noise is before putting in a coherent filter to eliminate it. At these low frequencies the propagation delay through the soil will be negligible so a coherent filter should be easy enough.

 

[Warpspeed]

Some very good points raised here gentlemen.

How about a fairly simple passive rough as guts front end to strip off all the really high amplitude impuse noise, followed by a narrow band active filter of some type ?

A commercial wave analyzer sounds like an excellent solution, provided there is power to run it where the measurements need to be made.