# How to design a cheap sine-wave generator up to 200 MHz?

I want to make a cheap wideband oscillator for an antenna analyzer I'm designing. I want a simple sine wave over a wide frequency range. I don't want to use a DDS IC like the AD9851 because it's expensive and feels like overkill.

I was looking at the SI5351A, which will generate a 50 ohm square wave clock up to 200 MHz.

I'd like to convert that square wave output into a sine wave over the range 1 MHz - 200 MHz. What is the simplest and cheapest way to do this?

Two ideas that come to mind are

1. Two cascaded op-amp integrators, using an OPA355 or something
2. A series of low pass filters that filter out everything but the fundamental, spanning the entire frequency range. For example filters with cutoffs of 2, 4, 8, 16 ,32, 64, 128, and 256 MHz? The correct filter would get switched to by an 8-port analog switch as the frequency rises. This seems like a lot of filters, but all these components are purely passive and would have relatively loose tolerances.

Does the approach of using a clock generator IC make sense? If so which of these filters make the most sense to convert the output to a sine wave? Thanks.

• Please be aware that you are basically building a transmitter and connecting it to an antenna. I have a vague remembrance of this type of antenna analyser being outlawed in Germany, but have been unable to find a source to quote. I remember it because back in the mid 1990s, a company I worked for had to buy an new analyser due to that law. Just saying, you might aggravate your neighbors and get visited by your country's equivalent of the FCC when you use that thing. – JRE Oct 28 '16 at 8:29
• Yeah, presumably the output would be in the microwatts. As you point out it would be incumbent upon the operator to make sure they were operating within bands they were legally licensed to do so. – bcattle Oct 28 '16 at 8:34
• How pure do you need the sine wave? What resolution in frequency? What resolution in amplitude control? What stability in both and ditto frequency jitter. If you want bottom end quality then your idea should work – Andy aka Oct 28 '16 at 8:48
• it seems like a digital clock would have excellent frequency stability? You can answer this yourself after reading the SI5351's datasheet. What is used as a reference for this chip ? Look in the block diagram and see where the PPLs get their reference frequency from. But I doubt this chip will fit your needs as it will very likely generate a very noisy (jittery) signal. It is designed to clock digital ICs. Digital ICs do not care about clock purity. For your application this might be more crucial. – Bimpelrekkie Oct 28 '16 at 10:11
• I agree with Andy that you must determine your requirements including your budget. Why are there no cheap 1 - 200 MHz sinewave generators for sale ? Because it is not that easy to generate a decent sinewave let alone for that whole range. If you want a decent (let's say less than 10% distortion) sinewave at a stable frequency then a DDS is the way to go. Up to 20 MHz a cheap DDS will suffice. 200 MHz is getting in the RF range so prices explode. Please prove me wrong by showing me a design on a budget that can do this because I have not seen any. – Bimpelrekkie Oct 28 '16 at 10:16

If you are prepared to use switched banks of filters you might as well consider using switched banks of colpitts sine wave oscillators. One transistor will get you a decent enough sinewave and add a couple of varactor diodes and you get a simple dc voltage control of frequency over a range greater than 2:1 i.e. one colpitts circuit gives you 100 MHz to 200 MHz (plus overlap with the next one down).

So, 8 transistor oscillators will do the job and sinewave purity will be better than about 5% I would say. This is my favoured colpitts oscillator configuration: -

I suggest you use a transistor with 5 GHz fT to get it working up at 200 MHz. The BB171 is currently available as the varactor and has a very good tuning ratio of 22:1. This tuning ratio implies a frequency ratio of $\sqrt{22}$ and that potentially is over 4:1 but you'll be very talented if you can engineer this range from a simple colpitts oscillator and get low distortion and amplitude stability.

To add a slice of quality you can feed the output to a HMC700 fractional-N phase locked loop and gain control of frequency and stability this way (using SPI); because you only have one oscillator selected at once, a single HMC700 should do the job for the whole range.

To select one of 8 signals can be done with pin diodes but it can probably done with less brain-ache using an RF analogue switch like the HMC544A. There will be others but you need to find ones with high isolation capablities.

You may also be able to use analogue switches to select a bunch of inductors that cover the whole frequency range - this would be an achievement because there will be stray and leakage capacitance issues but, the more I think about it, I reckon you could get at least a 5:1 range of frequencies from one colpitts oscillator and a couple of inductors switched in or out. This would halve the number of oscillators. Worth considering.

• This is a really neat idea, then yeah use the HMC700 to do closed-loop frequency control! – bcattle Oct 28 '16 at 22:15
• No huge isolation required of oscillator select switching if the unused VCOs are powered off when not selected. Then switching only has to isolate the poor impedance, not stop a signal. With care, you can magic up the bias for the PIN diode from the powered VCO, and reverse bias the unpowered ones! – Neil_UK Oct 31 '16 at 8:08
• @Neil_UK yeah, I thought about using the DC feed to each oscillator for that also but got way-laid and forgetful!! – Andy aka Oct 31 '16 at 8:11

Your second idea of using switched low pass filters to pass the fundamental of a square wave is the way it is done in many commercial RF signal generators. It does depend how clean you want your sinewave to be. It's quite difficult using an economical version of this technique to get better than 40dB typical, 30dB guaranteed suppression of harmonics, but that sort of level is adequate for many use cases.

There are several tricks you can employ to reduce cost and simplify design.

The first is to use filters at half octave steps, at least for the higher frequencies. Although a square wave has nominally no even harmonics, this breaks down for practical devices with asymmetries and breakthough resulting in a significant 2nd harmonic. At some suitably low frequency you can go to octave bands.

The next is to use low order elliptic designed filters, which improve the steepness of the transition band, at the expense of 'come-back' at higher frequencies.

The next is to arrange the casacade so that the highest frequency (so the one at which you are likely to have lowest power and lowest gain) goes through the shortest, least lossy, path, and that you add further sections as the frequency drops. A fixed well-designed 256MHz 'roofing' filter at the start of the cascade will deal with the come-back of the 192MHz filter, those two handle the 128MHz filter, and so on down.

The next is to switch the filters by passing current through PIN diodes, which is cheaper and easier than using other switching technologies. Bias current passes through the filter series inductors so biassing at a specific point in the filter cascade switches on the correct part of the filter, and the rest off.

The last is to only take the filters down to some reasonable frequency, and do the bottom frequency range in one go with a DDS and a single low pass filter.

• This is great, thanks a lot. The idea of using PIN diodes is really elegant – bcattle Oct 28 '16 at 22:11