Bandpass Filter for 1.6 MHz

Hello Hans,

Sounds like a low pass plus a high pass.

Come on, tuning the via slot of an inductor core isn't that hard. Sounds like you are German, in that case get a Bernstein tuning kit.

My clear choice would be LC. Unless cost is not an issue :-)

The topology depends on how true your group delays have to be. I suggest to buy a copy of William's "Electronic Filter Design Handbook". You could also design via SW. I used to do it with Otto Mildenberger's DOS filter tools. You can also design active filters with it if you really want to. Basically you enter the desired stop band, pass band and slope windows and it spits out filter order and values.

Regards, Joerg

the [...]

I'd definately go with Joerg on the LC's. The 1M6Hz filter (especially) with it's Q of 16 would be useless in that single opamp bandpass form. If you went L+C with a 10kohm source and load resistance then each filter circuit has a sufficiently low capacitor value that a 50pF trimmer cap' could provide the final tune. For inductors, I'd just use standard E12 Toko items. -Then- use a times 2 opamp to return to unity gain. john

Usually, one uses a low-pass filter after a DDS. If you need a bandpass, LC filters work well at your frequencies of interest.

If you really want to use active filters, consider using a gyrator based filter. Gyrators are quieter than a MFB or state variable filter. Plus, the component sensitivity of a gyrator is around 1. The state variable types tend to have a component sensitivity near the Q value. At your frequencies of interest, you will have GBW issues to deal with. See Jim T's web page for some gyrator ideas and hint:

If you have screwed up your DDS so badly as to require (apparently serious) cleanup you deserve to Flunk of be fired whichever is applicable. There are obsolete parts (over 12 years old) that would not engender this level of problems.

There are applications that would enjoy reduced level of out-of-band spurs below the DDS noise floor. It's not necessarily ignorance that makes an engineer look for standard design elements to achieve stringent design goals.

Since the only noise source of interest is the master clock and its harmonics, I'd suggest some (frequency-) distance to the clock, and lowpass the master clock out. So, have at least 16MHz Masterclock, this gives you a decade and thus 20dB per order of the filter.

Rene

Hello Joerg Thanks for your reply.

Yes, for the bandpass having 200 kHz center frequency a LP/HP combination would do the job. But for the 1600 kHz filter the Q of

I would rather electronically tune a varactor than use a trimmable inductor. Since money and board space is not critical, the additional tuning DAC per filter may be worth it.

I agree. Cost is not an issue though. Harmonics produced by the OpAmp would be a killer anyway. The goal is to obtain a REALLY low noise and spurious free output tone.

Another specification is that the delay be as short as possible. If the DDS output amplitude is changed, the filtered signal should react as fast as possible, i.e. in less than 10 us. This conflicts with the bandwidth requirement. Short group delay has priority.

Thanks for these suggestions.

Regards, Hans

Sali Rene

Besides the noise sources you mention, there is also quantization noise which sets the DDS noise floor, and some very weak spurious tones, no matter how clean the DDS clock is. The signals will be used for AC-biasing of some very sensitive sensors. Without filtering, the very broadband noise and together with the spurious tones would sum up to a power level comparable to the signals to detect. Filtering the DDS signal is therefore essential.

Regards, Hans

Hi Mark Thanks for the suggestion to use Gyrators. I will have a look at it. Hans

Hello Hans,

You could still do it with LP/HP if keeping the same architecture is key. Considering your group delay tracking requirements it might be key.

You may also need to generate low noise 10V-20V for the varicap unless you already have that.

Can't you just replace all the DDS with some nice low noise PLLs and be done with it? If you do that right you may not need any filtering.

Regards, Joerg

"Joseph2k" schreef in bericht news:T0XFf.1797$ snipped-for-privacy@newssvr27.news.prodigy.net...

serious)

Oh, and what part would that be? There is not much you can 'screw up' on a bare DDS itself, it's the DDS principle that is already screwed up, unless you believe that when your signal shown on an oscilloscope is a clean sinewave when it looks like that. There's all kinds of rubbish in it.

If quantization noise is the issue, then you need more bits on your DDS.

There are test-equipment-grade synthesizers (some of them do more than just sine waves) that promise 16 or 18 bits (or maybe more!) of amplitude resolution. A lot more expensive than a $5 chip from AD, in fact they are closer to $50000. I would think that seeing as how your max frequency is 1.6MHz (and these synths are often rated to hundreds of MHz) that there is a happy middle ground.

Tim.

As an afterthought, rather than constructing one tunable filter, it is almost certainly better to make a number of fixed-tuned filters and select the output via a multiplexer etc. You mention sweeping etc., obviously you do not want to switch filters in the middle of a sweep. Depending on switching timing, the multiplexer could be relays.

If you didn't have the 10usec sweep or slew time you mention elsewhere in the thread, this would be the perfect opportunity to use slug-tuned coils tuned by steppers or maybe a voice coil :-). (Don't laugh, it's a powerful technique.)

Tim.

Hello Tim,

I had a slug tuned oscillator a long time ago. It needed regular oil changes for the thread. Clean the old oil off until spotless. Then add weapons grade lube. Not too much but also not too little. Maybe a tad, or a smidgen, or a wee bit. This was almost an art.

Regards, Joerg

I'd go with your basic LC tuned circuit. It's hard to beat the dynamic range and noise performance of one or two LC's.