I'm working on a PIC-based telephone call costing device, that
indicates the cost of calls made by recognising the number dialled,
looking up the time per unit for that dial-code in a table, using the
correct call rate for the time of day, and timing the call.
(Why? The devices hired by our telecoms company that rely on metering
pulses are expensive & there's a long waiting list for them).

I'm having major difficulties reliably establishing when the call
actually starts - i.e. when the chap at the other end answers.
Sometimes, the line current reverses when the call is answered, and I
can use this. But for the other cases, I'm looking at the ringing
tone and seeing when it disappears, & assuming the call has been
answered.

I'm using the CML FX633 Call Progress Tone Detector chip at the
moment to do this - it indicates if there's a valid tone (ring,
engaged, etc) on the line - (you have to check the cadence in the uP
software). But it's too expensive here to make the project feasible.
I've tried to build my own filter (400Hz), but without much joy. It
definitely lacked the noise immunity & gain control of the i.c.

Can anyone offer any advice?

Many thanks & happy new year.
Rob.
*************************************************************
* Save a fortune on international calls, and earn commission*
* on people you sign up. Ask me how.                        *
* Rob Bruce-Brand      Independent Telephone Consultant     *
* bruce-brandra@durban.gov.za                               *
*************************************************************