FSK Radio design

ugh

e

our

, s

As has been mentioned before, the frequency shift is absolute - it is still 300kHz after the nominally 1GHz carrier has been mixed with a

990MHz local oscillator to get your - nomimally - 10MHz IF.

The problem with using a PLL is that the loop has to track a rapidly varying phase - when the signal is running at 10.0003GHz, the phase advances fairly rapidly with respect to the 1GHz nominal frequency, and when it is at 9.9997GHz it retreats equally rapidly. Your phase- locked loop has to have a loop bandwidth appreciably wider than 1MHz to track it, and you need at least a third-order tracking loop to track it particularly accurately - Floyd M. Gardener's "Phaselock Techniques" spells out the details.

The standard FM receiver uses a slow frequency-tracking loop to keep the local oscillator tracking the average frequency coming out of the FM signal source, but detects the frequency difference from that tracked average, rather than the absolute frequency shift.

The 4046 PLL chip includes a logic-based "phase" detecdtor, which doesn't actually detect phase, but frequency difference, and thus avoids some of the catches of using a clasic phase detector. Floyd M. Gardner is informative on this point too.

-- Bill Sloman, Nijmegen (but in Sydney at the moment, admiring the view of Sydney Harbour - including Fort Denison, but not the Opera House - from the living room of our flat).

I like that inductance circuit - it looks like it has the advantage over the circuit given here:

- that its maximum impedance isn't limited by the parallel resistance R connected to the op amp's noninverting input.

In ideal mathematical terms, mixing involves multiplying the incoming signal with a sine wave. So if your incoming signal is

cos(w1 * t)

and your LO is

cos(w2 * t)

then your result (you do the trig to verify)

cos(w1 * t) * cos(w2 * t) = 1/2 cos((w1 + w2) * t) + 1/2 cos((w1 - w2) * t).

So if w1 happens to be varying by +/- 300kHz before the multiplication, it'll still be doing so afterwards. But now you've got a signal component at w1 + w2 (which you usually want to throw away with a decent filter) and a signal component at w1 - w2 -- which you usually want to keep, and call your intermediate frequency signal.

You will need an IF filter, yes, but you need to choose a frequency scheme such that any signal that you don't want that may show up at the antenna can't get mixed down to IF. So if you want to receive exactly

1GHz with a 990MHz LO, you need to worry about receiving signals at 980MHz (because 1GHz - 990MHz = 10MHz, which is good, but 980MHz - 990MHz = 10MHz, too, which is bad - unless you can filter out the signal at 980MHz).

Beware of your 'bandwidth' rating -- the filter may pass out to 40MHz, but not block until 80MHz.

Kewl. Not that the part is not an I/Q downconverter -- just a simple mixer. So it'd be suitable for a superhet, but not for a direct conversion scheme.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

The SRQ2116ZSQ would do your conversion down to baseband, but it needs an external LO. Life ain't easy...

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

What is the insertion loss for the antenna side SAW filter ? This is directly added to the front end noise figure and hence directly degrades the system sensitivity.

Ideally, you should minimize any losses (cable and filter) before the first LNA and all filters have losses. In the current hostile RF environment, with myriads of RF sources, some means are required to protect the first LNA, by keeping most of the RF spectrum out of the LNA.

The loaded Q of the filter resonator will determine the bandwidth. Unfortunately, in order to get a narrow bandwidth, the resonator should be loaded lightly, but unfortunately this will cause larger attenuation within the passband.

For low passband attenuation, the unloaded Q of the resonator should be at least a decade higher than the loaded Q. Unfortunately, the 1 GHz is a problematic frequency for traditional helical resonators (quite small) or strip line resonators (too large), so modern alternatives would have to be selected with sufficient high loaded Q (and thus narrow bandwidth) without too much attenuation.

The filters after the LNA(s) can have large loaded Q and also high losses, since the noise figure is determined by the LNA(s) in a proper design. The post LNA filter will protect the mixer for out of band strong signals. For a single mixer superhet, the post LNA filter must also filter out the image frequency band. If image rejection (IQ) mixers are used, the up to 40 dB attenuation of the image frequency band might be sufficient and hence the image frequency band may be allowed to pass the post LNA filter (allows broader filter or using a lower IF).

If a PLL is used after the down mixer, some moderate filtering may be required to prevent the PLL locking on signals mapped to the harmonics (and subharmonics) of the IF.

A Q of 1000 at that frequency is rather low...

No doubt if you are talking about the unloaded Q for a cavity resonator or 1/4 wavelength stripline resonator. The problem at 1 GHz is that the free space wavelength is 30 cm.

The silver coated cavity resonator is quite big and even with a 1/4 strip line bent into U-shape will require several centimeters in the longest dimension. With a suitable dielectric, the size could be reduced further, but the dielectric losses will increase.

While large filter dimensions are not a problem for a large, bulky, high power transmitter, but for receiving applications, the bulky filters would be unproportionally large, compared to the actual electronics.

C that varies with frequency. Lowpass and slice. It's just a >classic discr= iminator with a high-Q resonator.

When you build a discriminator the amount of absolute delay in the delay device is a critical part of the design. How much delay does the delayed path have? It needs to be matched to the data rate right? How does this design hold up to DME interference that is 1MHz off and

30 dB higher than the signal you are trying to demodulate? You have proposed a nice MacGiver solution but we both know that it is not a practical approach to solving the OP's problem.

that varies with frequency. Lowpass and slice. It's just a >classic discriminator with a high-Q resonator.

A single-pole resonator has delay proportional to Q, and Q is easy to kill. Getting 1 MHz data out of a 1 GHz carrier isn't a challenge.

DME?

The OP said he has two SAW filters ahead of the discriminator. It's their job to shape the bandpass, not the discriminator's job.

You have

A coaxial resonator and a mixer is a perfectly viable discriminator.

John

Essentially all FM detectors have the same behaviour at high SNR--it's the low-SNR case that's hard work. Have you done a link budget?

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics Electro-optics Photonics Analog Electronics

55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

There used to be a lot of 900 MHz chips available for this sort of job--I did a 900-MHz frequency-hopping spread spectrum link a dozen years or so back (it was part of the ad-hoc network for Footprints). None of the chips I used are still available. You might find some holdouts.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics Electro-optics Photonics Analog Electronics

55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

e DC that varies with frequency. Lowpass and slice. It's just a >classic di= scriminator with a high-Q resonator.

It is a perfectly viable discriminator, but not a perfectly viable receiver. BTW- The SAW filters (at 978MHz) are used as preselector filters , they are not used to achieve proper selectivity. An IF filter would be used for that. You know this right? I mean come on...proposing a discriminator at 978GHz to directly demodulate an FSK signal for a practical receiver?

I gather you didn't bring Gardner with you, because he doesn't say that. ;)

It isn't hard for a PLL to track rapid modulation if its bandwidth is nice and wide. 1 MHz isn't that fast for a feedback loop.

Third order loops (i.e. three integrators in the loop plus a couple of zeros further out to make the loop stable) ideally have zero phase error due to a frequency ramp. Back in the day, that helped with stuff like Pioneer 10, whose Doppler changes slowly with time. It's entirely irrelevant for a FSK demod, where df/dt goes all over the place, and you only need to distinguish a mark from a space with low bit-error rate. It would also be hard to keep it stable in the presence of fading and such stuff, which can make the loop gain vary a lot.

The 4046's PD2 has a frequency-phase characteristic, i.e. after filtering, it's a linear phase detector for phases between -pi and pi, but sticks at the limits (it more or less rails) so the loop pulls in from anywhere in the lock range.

In any event the OP is describing a very narrowband FM system, with a modulation index of only 0.3 (300 kHz delta at 1 MHz f_mod). The modulation index is numerically equal to the maximum phase deviation in radians, so the peak phase deviation is only 20 degrees or so.

The OP would probably be better off using a PM-type demodulator, i.e. a narrowband PLL to track the carrier, with the output taken from the phase detector instead of the VCO input.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics Electro-optics Photonics Analog Electronics

55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

Helical resonators work up there, and have Qs high enough to make a discriminator around a GHz. They are small and cheap.

John

On a sunny day (Fri, 24 Jun 2011 10:41:15 -0400) it happened Phil Hobbs wrote in :

For 900+ MHz I would use an old TV tuner with PLL, get the 38 MHz or so IF, do the TBA120 sort of thing (FM detector) at about 35 MHz, or mix it down again to say 1 MHz, filter, and use a one shot and lowpass, or any 4046 PLL. There are many configurations that could work. You can then tune all the way in that 5 MHz wide IF band.

but I am

y.

e l

Tim, That is extremely helpful. I see your point on the image frequency considerations, it seems that the entire bandwidth presented to the mixer has to be taken into account. It does not seem that it is important, in this case what the IF is, I could pick any value. How about increasing the the IF to a value that will push the image frequencies outside some range with the range being determined by a IF filter bandwidth?

Rich

that varies with frequency. Lowpass and slice. It's just a >classic discriminator with a high-Q resonator.

"Practical" depends on his application. I don't know if he is receiving satellite data or if he is building a link to run 50 yards. Deep-space receivers are one extreme, garage door openers are another.

A coaxial or helical resonator could have a Q of 1000, namely a bandwidth of 1 MHz at 1 GHz. That could make a relatively narrowband discriminator. Heck, use two of them, and make a TRF receiver. Helicals are tunable, and come in duals, which might be interesting.

My suggestion is that, since stable high-Q resonators are available at his frequency, discriminating up there might work.

Ideas shouldn't be shot down without playing with them a little first. Usenet is a terrible place to brainstorm ideas, because there will always be people eager to kill the process.

John

that varies with frequency. Lowpass and slice. It's just a>classic discriminator with a high-Q resonator.

Where's the fun if everyone thinks it's easy? (Working hardware settles a lot of arguments.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics Electro-optics Photonics Analog Electronics

55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

Good point. Standard cheap teevee parts could solve 90% of his problem. Maybe filter and discriminate at the first IF, using standard resonators.

A second mixer would be easy, too.

John