Tunable Bandpass Filter

Couple of questions:

-- Is the entire analog channle also 2-7MHz? Or wider?

-- What power levels are you dealing with?

A few approaches are:

-- Mix your signal with an LO of, say, 19.4-14.4MHz such that the band center of interest is at 21.4MHz (use a low pass filter so that you don't pick up the image frequencies above 7MHz). Use a cheap off-the-shelf 21.4MHz IF filter (probably ceramic) to get your 600kHz passband (this is a Q of

21.4MHz/600kHz=36 -- easy peasy). Mix again with the same LO to put your center band back where it came from. (High power levels -- much above, say, 0dBm -- start creating intermods and compression problems from the mixers.)

-- Build yourself a bank of switched capacitor and inductors that get switched in and out as appropriate to "build" a bandpass filter wherever you need it. (Use PIN diodes or MMIC switches for the switching.) If you need very fine control you'll end up using a varactor diode (or perhaps a DC bias on an inductor) to set the exact center frequency. (High power levels here push your varactor or inductors far enough outside of their linear ranges that get start getting frequency responses that are functions of power levels as well as intermods.)

-- Same as above, but use relays for switching inductors and capacitors in and out and motorized variable capacitors (or slug-tuned inductors) if you need fine tuning. (Higher power levels are attainable, but you end up consuming a lot of physical space and tuning is slow.)

If the filter is simple enough, you *might just* be able to get away these days with an FPGA-based "all digital" implementation: Feed your signal to an ADC, have the FPGA run a FIR or IIR filter, and spit it back out to a DAC. As with most things "DSP," there are a lot of upsides, although your signals are at a high enough frequency you'll probably consume a fair amount of power running all the multipliers in your FPGA, and it isn't going to be the "bargain basement price" series of FPGAs that'll have enough horsepower to pull it off.

---Joel

On a sunny day (Mon, 16 Nov 2009 12:10:17 -0800 (PST)) it happened Guy Eschemann wrote in :

You could mix up to some higher frequency, use a fixed filter at that frequency, and then mix down again. And tune the local oscillator.

It would help to know how sharp you need the filtering.

How about visiting a library and reading some relevant books?

Chris

Joel,

There are 8 non-overlapping analog channels in the range between 2 and

7 MHz. Each channel is approx. 600 kHz wide.

I'm not sure about the power levels yet, but the channel selection filter comes after the preamplifier and the receiver main amplifier (AGC), so the amplitude is pretty much controlled at this point.

If possible, I'd like to avoid any mixing up and down. I'm actually considering a mixerless approach (bandpass sampling) to translate the channel of interest down to DC, so it would be really annoying to mix the signal up and down just for filtering.

Also, I don't want to use a digital filter at this stage. This would require sampling the band of interest at something like 30 MHz, which is bad for power consumption.

Thanks for your help! Guy.

nter

p the

lter

your

say,

.)

tched

it.

fine

push

get

ell

n and

ed

ming a

e

an

. =A0As

are

o

I remember seeing an article years ago about a dual mixing technique that used two minicircuits sbl1 type balanced mixers in cascade with a common variable local oscillator and a crystal filter interspersed between the two mixers. Draw it out on paper and do the sum / differences to see how it works.

You don't say how many db/octave at the edges, but a crystal filter can provide a very sharp rolloff. The 600Khz passband may be better handled by an lc or active filter with fast opamps, depending on requirements.

Otherwise, how about using fast opamps in a byquad or state variable configuration ?...

Regards,

Chris

I'd love to hear it if you could point to any book that has a large amount of text specifically devoted to *tunable* filters. I have plenty of filter books (including many of the "classics"), and most give little more than passing mention to them. (I suppose because -- other than the "mix it up to a fixed frequency with a good filter" method than Jan and I mentioned -- most implementations I'm aware of are some variety of the "brute force" method anyone would think of, so perhaps there's not a whole lot to say...)

One approach I forgot to mention: I have seen people build active filters with multiplying DACs as the tuning elements up to better than a MHz, but I think

7MHz would be quite a stretch (the DAC's parasitics start to eat you alive).

I've messed around with gyrators occasionally, and while you can build them to tens of MHz with fast op-amps, tuning is still tricky -- the last time I went down that path I convinced myself a way to make it work might be to bulid a set of two filters with matched tuning elements, have one be the "real" filter, and the other servoed to it via its twin that's constantly seeking to peak a synthesized signal (from a DDS or whatever) that's going through it. Alas, this approach is best for small signals and an IC implementation.

Speaking of which... the IEEE has plenty of articles on tunable filters, but most are oriented towards IC implementations. Too bad the standard membership fee of $176/yr (!) doesn't get you *any* on-line access to the them...

---Joel

Ideally: -60dB within the 80kHz guardband.

For 8 channels I think I would just build 8 lumped-element filters and switch them in and out. -- It's easier to get decent repeatibility using, e.g., 2% inductors and capacitors when each component only influences one frequency band rather than a single "switching" filter where multiple inductors and capacitors may interact and their tolerances then build on one another.

Depending on what order filter (how sharp) it needs to be, 8 filters can still be pretty compact.

---Joel

On Mon, 16 Nov 2009 13:19:33 -0800, Guy Eschemann wrote: (top posting fixed)

center

up the

filter

your

say,

mixers.)

switched

need it.

fine

push

that get

well

in and

need

consuming a

these

to an

DAC.  As

signals are

power

to

Joel,

There are 8 non-overlapping analog channels in the range between 2 and

7 MHz. Each channel is approx. 600 kHz wide.

I'm not sure about the power levels yet, but the channel selection filter comes after the preamplifier and the receiver main amplifier (AGC), so the amplitude is pretty much controlled at this point.

If possible, I'd like to avoid any mixing up and down. I'm actually considering a mixerless approach (bandpass sampling) to translate the channel of interest down to DC, so it would be really annoying to mix the signal up and down just for filtering.

Also, I don't want to use a digital filter at this stage. This would require sampling the band of interest at something like 30 MHz, which is bad for power consumption.

Thanks for your help! Guy.

You're kind of painting yourself into a corner, and the corner is called "eight non-overlapping band-pass filters".

Or consider that you're already on the wrong side of some active electronics, so you've already levied most of the disadvantages of a superheterodyne receiver against yourself. Why not just go the rest of the way and make it a superhet? Upconvert to something convenient like that 21.4MHz, filter, then use a fixed downsampling scheme.

The whole reason that Armstrong invented the superhet was to dodge the difficulty of trying to make a good agile filter at RF -- here you are 75 years later struggling with the same problem, yet the answer may still be the same one.

--
www.wescottdesign.com

You need a superhet.

--
www.wescottdesign.com

and a crystal filter, unless you mix down to baseband + a low pass filter.

There are probably mobile radio ic's that would do the job, but not really my field. Silicon Labs, Broadcom and others...

Regards,

Chris

Thinking again, plug 'software defined radio' into google...

Regards,

Chris

Yea, but a very real problem that SDRs have is that while, sure, you can get beautiful, near-vertical skirts around a filter, if there's a strong interferer nearby, you have to filter it prior to digitization or at best you lose SNR for the intended signal (desensing)... and at worst that SNR goes negative!

Although you probably know this. :-)

Baseband + lowpass will be difficult. Normal I/Q downconversion with the carrier "off to the side" probably won't give you 60dB of rejection. You could get that rejection by using the Weaver method, but only if you could tolerate the DC bias in your baseband signal squatting right in the middle of the signal that you're trying to decipher.

OTOH, if you _could_ deal with the DC bias, then it may work quite well, and fit well into your preferred downconversion scheme.

--
www.wescottdesign.com

From analog radio days, yes, but have no experience of sdr at all. Just something i've been reading about in the last few months. From what I can see, it doesn't get round the need for a low phase noise lo to prevent filter washout, irrespective of how the filter and processing are implemented.

A crystal filter is still hard to beat on cost and performance, even now :-)...

Regards,

Chris

For ultrasound engineers and Radar guys it's routine, except that we call them tracking filters. They consist of a fixed lowpass and a highpass that's tuned downwards while echoes are received. The challenge is to make them reproducible in production without any alignments. Many tricks there, such as servo or pilot tones, but that's as far as I am allowed to speak in public.

The challenge with that would be to find matched tuning elements. It'll be almost down to PIN diodes which can be had as duals and dual FETs but those aren't so hot when you need a spiffy shape factor because they are resistive elements.

Then there's the old scheme of having several resonant filters spread apart like it was done in the old tube-era TV sets. 2-7MHz is feasible but that's close to the reasonable limit. 600kHz BW at 2MHz will be a real stretch with a resonant scheme.

Quite frankly, I'd consider a DSP here.

Don't get me started on that ... it is the reason I have stopped writing for IEEE.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

books

[snicker]

Probably the same way I do sonar ;-) ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |

If you wanted a President with balls why didn\'t you elect Hillary?

The statement 'The middle frequency is in the range 2..7 MHz, and the required bandwidth is 600 kHz' suggests something similar to the front end of an HF or VHF receiver, about which a great deal has been written. I don't have my own library to hand (I use the IET in London) but I daresay Google knows the names of some the relevant books

. Of course, a great deal of filtering (!) of the resulting 978 returns is then needed, but that would be a task for the OP.

I think the statement 'How do I get started with this?', in the absence of any evidence of effort spent learning the basics, deserves the the response I gave.

Chris