Interactions of Voice Band Data Modems with Network

Interactions of Voice Band Data Modems with Network Echo Cancellers Bob Reeves BT Issue 1 21 April 2010

Overview This presentation will cover problems encountered with two different types of low speed data modems and their interaction with network echo cancellers: • • • • V. 23 telemetry modems used by the UK Water Industry to monitor lakes, reservoirs and inland waterways V. 22 bis modems used in Automatic Teller Machines (ATMs) and Electronic Point of Sale (EPOS) terminals Both of these problems are caused by the echo canceller’s Non-Linear Processor (NLP) Good opportunity to encourage EC designers to follow guidance in ITU-T Recommendations with respect to NLP design © British Telecommunications plc

V. 23 Telemetry Used by UK Water Industry to monitor water levels in lakes and reservoirs. Remote out-stations report information to central in-station over dial-up connections EC EC Network (PSTN) Out-station Modem EC Out-station Modem In-station Modem EC Out-station Modem Remote out-stations often on long lines © British Telecommunications plc Out-station Modem

V. 23 half duplex modulation used for telemetry • Half duplex V. 23 (FSK) at 1200 bit/s • In-station modem sends a request to the outstation modem • Out-station modem responds very quickly (in the order of 15 to 30 ms) • 2100 Hz answer tone may be present at the start of the call, but plenty of silence to allow NLP to re-enable • Example V. 23 telemetry call Turnaround = 15 ms © British Telecommunications plc In-station Request Top Trace recorded at In-station (2 -wire) Network Delay Out-station Response Bottom Trace recorded at Out-station (2 -wire)

Example of an unsuccessful V. 23 telemetry call • In-station modem sends a request to the outstation modem • Out-station modem responds very quickly (in the order of 15 to 30 ms) In-station Request Truncation of signal by NLP 30 -40 ms • Out-station response is truncated or clipped as it passes through network echo canceller • Dependent on line length (long lines cause failures) • Problem isolated to NLP by manually disabling the NLP. This resulted in successful calls Turnaround = 15 ms © British Telecommunications plc Top Trace recorded at In-station (2 -wire) Network Delay Out-station Response Bottom Trace recorded at Out-station (2 -wire)

Breakdown of an Echo Canceller Adaptive filter forms model of echo path to “cancel” echo. Note that in this example ONLY the reflection from the out-station side is cancelled. Echo Canceller Adaptive Filter Comfort Noise Generator inserts noise in place of background noise when NLP is active In-station modem side CNG NLP Non-Linear Processor removes any “residual” echo after cancellation. Acts as a suppressor. © British Telecommunications plc Out-station modem side + HP Filter High Pass Filter removes any DC component from the echo path Echo

What can we do about it? Network: • Re-design NLP with faster de-activation time (transitions 2 and 3 in G. 168 Figure 39) – preferred longer term solution • Use separate V. 23 detector - turn off NLP before it gets the chance to clip the waveform – acceptable work-around • Increase line card gain to remote sites where failures occur (although many sites so remote that they are already at their highest gain setting) – not an option in most cases and would result in “special” treatment for particular lines with associated long term overheads Protocol: • Add redundancy (null characters) to initial out-station response so that clipping has no effect – not an option in practice since it places the burden on the customer to modify in some cases 1000 s of remote units © British Telecommunications plc

G. 168 Figure 39 – NLP operating regions © British Telecommunications plc

G. 168 Annex B – Reference NLP © British Telecommunications plc

V. 22 bis ATMs & EPOS Terminals Used by cash machines (ATMs) and for Point of Sale transactions in shops, restaurants, etc. The ATM or EPOS terminal uses V. 22 bis to complete a transaction over dialup connections EC EC Network (PSTN) ATM EPOS Terminal EC ATM Central Modem EC EPOS Terminal © British Telecommunications plc

V. 22 bis full duplex modulation (ATM) • Full duplex V. 22 bis (QAM) at 2400 bit/s • Analysis performed in frequency domain No Answer Tone Central Modem (Answer) • 2100 Hz answer tone not always present at the start of the call • Example V. 22 bis ATM call © British Telecommunications plc Captured on 2 -wire point at ATM S 1 Signals Unscrambled binary 1 s ATM Modem (Calling)

Example of an unsuccessful V. 22 bis ATM call • 2100 Hz answer not present at the start of the call • S 1 signal is not recognised by Central Modem which tries to connect in V. 22? No Answer Tone No S 1 Signal Central Modem (Answer) • ATM cannot fall back to V. 22 so call fails • Dependent on line length • Problem isolated to NLP by manually disabling the NLP. This resulted in successful calls • Example unsuccessful V. 22 bis ATM call © British Telecommunications plc Captured on 2 -wire point at ATM Unscrambled binary 1 s S 1 Signal ATM Modem (Calling)

Failure Mechanism • Difficult to establish the exact failure mechanism here • We know that turning the NLP off in the echo canceller facing the ATM or EPOS terminal cures the problem • Dependent on line length (long lines cause failures) • Truncation of S 1 signal from ATM or EPOS terminal by NLP the suspected failure mechanism (but not proven) © British Telecommunications plc

What can we do about it? • Re-design NLP with faster de-activation time (transitions 2 and 3 in G. 168 Figure 39) (assuming failures are due to NLP truncation) – preferred longer term solution • Detect V. 22 bis modulation (unscrambled binary 1 s) and turn off NLP – acceptable work-around • Increase line card gain to remote sites where failures occur – would result in “special” treatment for particular lines with associated long term overheads • Works if Answer Tone present, since echo cancellers turn off NLP on detection of 2100 Hz © British Telecommunications plc

Summary & Conclusions • • Problems encountered with two different types of low speed data modems and their interaction with network echo cancellers: – V. 23 telemetry modems used by the UK Water Industry to monitor lakes, reservoirs and inland waterways – V. 22 bis modems used in Automatic Teller Machines (ATMs) and Electronic Point of Sale (EPOS) terminals Both of these problems are caused by the echo canceller’s Non-Linear Processor (NLP) • Some echo cancellers do not exhibit the problem so it is possible to design an NLP that does not interfere with these modems • Echo Canceller designers are encouraged to follow the guidance in ITU-T Recommendations for NLP design, especially G. 168 Annex B and the target timings given in Tables B 1 and B 2 © British Telecommunications plc