DFT taps for DTMF

Hmm, no idea mate..... we still have to sort out the top versus bottom posting debate around here.

Anyway, why are you talking about taps?

Whoops, my cake is burning.

Bye

DNA

Hello Roger,

You could match the BW to the old standard. Don't remember where that was published, could have been Bell. Nowadays it'll almost always be more precise in frequency but that depends on the application, whether or not there is a chance that signals could come from an old phone. Also, there are regs about how fast you must be able to detect each tone event (20msec?) so you can't go too low in BW.

There's a very clever DTMF DFT at

If you search for "Goertzel algorithm" and "dtmf" you'll find what you're looking for. I've used this algorithm several times with good success. There is additional processing needed to meet certain requirements, but without knowing your exact needs it's tough to recommend a complete solution.

Bob

I found using 16 samples was a good compromise. This gives filter ranges with little overlap but still decodes each quickly. The method decodes all 8 frequencies at the same time by scheduling 'convenient' times to take each sample.

Luhan

The frequencies not being critical is one of the problems for detecting DTMF. You'll need to process the signal in a way you know the frequency of the incoming signal precisely. There are extensions to DFT which use a sliding window and the phase information to extract the dominant frequency in an piece of the frequency spectrum.

QDFT? That is brilliant, and not just for DTMF. There was a project I did with sub-tone signals some years ago that could have been done with cheap micro rather than the DSP I used.....had I thought about this technique. I suppose it would work anywhere were you can control your sampling.

BTW, you only need to look for seven frequencies because the 4th column is never actually used on phones.

If there is a need to have the DTMF detector enabled while there's a possibility of having voice energy present then you should look at 14 frequencies -- the 7 row and column frequencies plus each of their second harmonics. The idea is to make sure that there is no second harmonic energy if the fundamental is detected. This technique works pretty well, to not detect possible voice energy as DTMF, because voice is rich is harmonics. For example, if the detector senses 697Hz and also sees 1394Hz then it is most likely voice and not DTMF. Search for the term "talk off" and "DTMF" to see what this is all about.

A properly-functioning DTMF decoder is much more complex than merely doing some sort of digital filtering to detect row and column frequencies. It all depends on what the application is whether or not you need to make it more or less rigourous.

For example, we once used a voicemail system that while speaking its own menu would detect this outgoing voice as an incoming DTMF digit (depending on the selected trunk's 4 wire transhybrid return loss). It turned out that the dectected digit meant "delete message". It was pretty comical, and yet very annoying if you didn't want that particular message deleted.

Bob

Just realised, you are the "inventor". Congratulations :-)

energy

to

all

that

Yes, the well known issue of simplifying just a little too much; and a surprise gotcha.