AM radio receiver - design

Why vertical polarization? Most man made noise is vertically polarized.

A high Q preselector will help with SNR.

Many have made some good suggestions on the antenna front.

For a receiver, if you really want something economical that can pick out exactly the bandwidth of interest and reject everything else, try looking at a quadrature sampling detector, aka Tayloe Mixer. The basic idea is to do a very high dynamic range commutating mixer with four phases of a digital clock, giving you a a quadrature IF either at zero or in the kilohertz range. You use some op-amp filters to remove strong interference, then put the result into a high quality stereo sound card (it's still an RF signal at this point not 'audio' but the frequency range is within the bandwidth of high end audio converters). From then on, the remainder of the signal processing is all digital - highly customizable IF filter bandwidths, detector algorithms, etc.

And AM broadcast stations. :-)

Hope This Helps! Rich

I can't explain why Outlook Express isn't inserting '>' characters at the beginnings of lines, but I will precede each of my comments by @@ below:

- - - -

I think I was completely clear in the first posting, but that you jumped to assuming something because you were not thinking about the OP's case. The OP is making a radio for the standard AM band.

@@ I had never lost sight of that.

Keep that point foremost in your mind as you reread what I wrote back there and read this explanation below. They are two different cuts at trying to explain the same ideas.

Disregarding signal strength, the signal in a normal AM radio is ideally:

Y = f(t) * sin(wt) + sin(wt) (1)

Where:

f(t) is the program material that is centered around DC. It contains frequencies from 20Hz to roughly 7KHz. The peak amplitude is always less than one.

sin(wt) is the carrier. It appears twice because this could also have been written as:

Y = (1 + f(t)) * sin(wt)

The version (1) of the equation is the better because it breaks the carrier and the sidebands apart and makes it far more clear what the signal is:

What you actually receive will be:

Y = fn1(t)*cos(wt) + fn2(t)*sin(wt) + f(t)*sin(wt) + sin(wt)

The fn1 and fn2 are two parts of the noise. Any function (noise or signal) in a band that does not include DC can be broken into two parts in this way.

@@ This does not represent noise in a band of frequencies, as is encountered in AM reception. Simply multiplying a function by sin and cos waves works fine for a single vector but does not account for the different Fourier components of the noise. There are many well-respected text books that provide the correct way to resolve noise into amplitude and phase components, and this always involves integrals.

Since you know single side band methods, I won't bother to explain how. Just remember to remind yourself that this is not SSB radio before reading the next bit.

Now that we have what the receiver actually has as a signal. The next step in the argument is to assume that we have a PLL that is locked onto the carrier and that its bandwidth is much less than 20Hz so that as far as we are concerned, it is a constant frequency sine wave. We can multiply each term in the equation by sin(wt)

fn1(t)*cos(wt) * sin(wt) no base band result

@@ Only true for a single vector, not for a frequency band of noise.

fn2(t)*sin(wt) * sin(wt) base band is fn2(t)

f(t)*sin(wt) * sin(wt) base band is f(t)

sin(wt) * sin(wt) DC result

Notice that only the fn2(t) appears in the result.

@@ ... as it would if we were dealing with a single vector.

Now for the case where we don't use a PLL. To keep it simple I will show only the second power term of the diodes function. The diode curve is actually a long series but I am too lazy to write an infinite number of terms. I guess it would be valid to say that I am using an RMS based detector. I will also not break down the fn1() and fn2() into series. Just remember they are. I also won't show the signs.

fn1(t)*cos(wt) * fn1(t)*cos(wt) fn1()^2 noise fn2(t)*sin(wt) * fn1(t)*cos(wt) no base band result f(t)*sin(wt) * fn1(t)*cos(wt) no base band result sin(wt) * fn1(t)*cos(wt) fn1() noise

fn1(t)*cos(wt) * fn2(t)*sin(wt) no base band result fn2(t)*sin(wt) * fn2(t)*sin(wt) fn2()^2 noise f(t)*sin(wt) * fn2(t)*sin(wt) fn2()*f() noise sin(wt) * fn2(t)*sin(wt) fn2() noise

fn1(t)*cos(wt) * f(t)*sin(wt) no base band result fn2(t)*sin(wt) * f(t)*sin(wt) fn2()*f(t) noise f(t)*sin(wt) * f(t)*sin(wt) Program material squared sin(wt) * f(t)sin(wt) Program material

fn1(t)*cos(wt) * sin(wt) no base band result fn2(t)*sin(wt) * sin(wt) fn2() noise f(t)*sin(wt) * sin(wt) Program material sin(wt) * sin(wt) DC

If you sum it all up, you will find that both fn1() and fn2() appear in the resulting function that goes into the squareroot if this is an RMS.

@@ I can only suggest that you consult a technical library and look up the correct way to represent a band of noise as orthogonal components. It would certainly require notation beyond the capabilities of plain text. Let me know via this thread if you'd like me to suggest some leads.

Chris

d.

Get a grip. The idea is to receive the service without interference. Using a Roku or Squeezebox to receive the audio is probably cheaper than what it takes to get around QRM/N. Since I own both types of equipment, I am qualified to make such a statement.

I'm amazed that has a patent. It is very similar to designs used at least a decade earlier.

In the early 80s, we did a modem using a similar scheme, but at that point it could be done in DSP. The easiest way to derive IQ signals via sampling is by decimation and alternate sample inversion.

Assume this is your sample stream

This assumes sampling at 4x. You could need to do clock recovery of the carrier with a multiplier.

Direct downconversion -> the receiver is hit by the 2nd order IM products -> poor dynamic range.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

It's all a terminology thing -- you and I know that people have been building mixers that effectively consist of nothing more than hard-saturated transistors or diodes for decades, whereas in recent times a lot of people went to school, and the professors there only teach about analog multiplier-type mixers where it's a nice, pretty sine wave going to the RF port (and of course the math is cleaner). Hence, someone gets the bright idea that... hey, what would happen if we did this "digitally" but just using some ideal switches? Surely that's better than building analog multipliers!? -- And effectively the mixer gets re-invented...

re

red

rks

Yes, it does apply to the noise in this case because we only care about the base band result and not the high frequency results. If we cared about the signals out of the demodulator at high frequencies, it would be quite a different matter.

Also note my comment about my being unwilling to write the infinite series needed for a full treatment.

This would be why the people attempting to make PLL based radios are having such poor results. The phase/frequency detectors have awful performance in a noisy environment. The noisy environment is where it is worth the bother of using a PLL.

This device would sample the 455KHz signal, so the signal has been mixed twice by conventional means.

HMMM... good point.... you are correct...

I recall now that you mention it, it is better to use a linear phase detector to recover noisy signals......

Mark

of

ame

time

Here is the paper:

First of all, if he was really serious about studying such issues, he would set up a waterfall display and prove the flutter occurs at the highest frequencies of the band. But no, he just proclaims this is where the flutter occurs. Then what is his solution? He just filters off the high frequencies. So really, I have stated the situation quite accurately.

Then look at that filter. I mean serious, discrete LCR with questionable source and load impedance? In any event, I could just switch in a tighter filter and get the same effect. Of course I did that, and it didn't effect fading in the least.

Then look at this one:

Believe what you want. Mr. Langford just has too much crap in his papers for me.

Yes, it does apply to the noise in this case because we only care about the base band result and not the high frequency results. If we cared about the signals out of the demodulator at high frequencies, it would be quite a different matter.

Also note my comment about my being unwilling to write the infinite series needed for a full treatment.

- - - -

@@ Both irrelevant. You don't have to believe what I say but you owe it to yourself to at least investigate the possibility that you have made an error, and look up the correct way to resolve noise components in a band.

I believe what is stated in a number of well-respected books, and is demonstrated in a small proportion, that there is no difference in the S/N provided by an envelope detector or a synchronus detector in the conditions we've been discussing, and you disagree. Your analysis hasn't convinced me that you're right.

Let's end the argument here - I don't suppose many others are interested.

Chris

he

ks

t to

I suggest that you do the same. Consider making one frequency in one side band:

Y =3D sin(At) * sin(wt) + cos(At) * cos(st)

This does one frequency. I can now do two frequencies in the same side band as:

Y =3D (sin(At) + sin(Bt))*sin(wt) + (cos(At) + cos(Bt))*cos(wt)

We can get the other side band by changing the sign from adding the sin () parts to subtracting it.

We can build up any collection of side bands we require in this way. If we are willing to do an infinite number of terms, we can get the noise as I suggested.

he

I may have zipped right by it but, yes, the multiplier type phase detector is best in the noisy case. The multiplier doesn't need to be a full multiplier. A gain stage that flips back and forth between a gain of one and minus one will do. Since the input signal is from a tuned circuit, the issue of harmonics doesn't apply.

Another subject I zipped right past is the reason that the VCO or reference is running at 2 times the frequency. Consider this circuit:

V1 V2 IN --+--/\\/\\--+--/\\/\\--+--+--/\\/\\--+--/\\/\\--+-- ! ! ! ! ! ! !4053 -!-\\ ! !4053 -!-\\ ! O ! >--- O ! >---

Wow, he is putting a large capacitor across the audio amplifier output. That should sound *interesting*

One of the problems with terminology is what the circuit really does.

You can stick a PLL on the BFO and feed a product detector. (Can growl at you). Or you can use a narrowband filter and a limiting amplifier to regenerate the carrier. (Can drop out if the carrier fades).

You can use both sidebands into a plain product detector, or you can use a image reject a.k.a. I/Q mixer with phasing to cancel out an unwanted sideband.

Various combinations end up doing different things.

Using both sidebands does a good job cleaning up signals with selective fading, where chunks of the audio frequencies in the sideband are getting canceled out. Common in signals coming in over the poles. Aurora flutter.

And using a image reject mixer will cancel out the signals on one side of the carrier better than the skirt response of the IF filter.

But they all get lumped under the synchronous detector label. The ones with the best reputation are the Sony receivers that use a chip designed to demodulate Kahn style AM stereo, using image reject mixer(s).

Actually, it's a smoking crater infested with right wing zombies. But there should be a lot of useful stuff in the archive. Ron Hardin? is (I think) the guy with a yard full of fancy antennas and noise phaser boxes. Another poster to look up is Pete Gianopolus [or something like that] who was working on a very high performance medium wave receiver last year or two.

Mark Zenier snipped-for-privacy@eskimo.com Googleproofaddress(account:mzenier provider:eskimo domain:com)

I know. He should have used the dual of the circuit. Series L rather than parallel C. Even then, the source impedance isn't known. I never bothered to see if the component values are correct. The whole implementation is stupid enough that I didn't believe checking the design was worth the effort.

I have little faith in Mr. Lankford's theories, but at least it keeps him off the streets. Granted, it sounds like he did use some cheap synchronous demod radios since he complained about the growl. As I said, sync is hard to do well in practice. Drake wouldn't have made the B rev if the sync in the older model didn't suck. The AR7030 probably had the best off the shelf synchro for a consumer radio, but even the designer complained about it.

See the bottom of this page:

Some other interesting reading about the design:

I never own a Sherwood sync demod, so I can't comment on the quality. In fact, I never met anyone that owned one. However, Sherwood does test a lot of radios and is considered well in the industry.