I’ve been tinkering around with VoIP gateways lately. After I got an old rotary dial phone working over the internet, using a free SIP service and a smartphone, I tried it “in reverse”, and offline: to establish communication between 2 computers and get them on the internet. And that worked!
But what if a proper modem was not in place? Will it work over a phone?
I only had one telephone – the server side uses a modem – otherwise I’d make 2 couplers 🙂
The phone is a 1950s Model “K” made by Tesla Prague; a Czechoslovak postwar copy of the Siemens-Halske W28.
Featuring the IBM PCjr as one of the test clients.
A Grandstream HT802 VoIP ATA in offline mode (static IP 127.0.0.1 with direct port calling) is used to emulate a landline, except for the beginning, where it is paired with a Linphone SIP server.
Well of course it would work! And that’s how it was done over fifty years ago.
The data transmission had to be slow enough to be turned into audio and accommodated for the limited bandwidth. And the first modems were wooden boxes, designed to fit a phone handset. Why? At that time, most telephone companies were monopolies, with phones and lines being their property. Connecting anything else was banned. As such, these devices were called acoustic couplers: they had no connection to the landline, and used their own speaker and microphone to temporarily attach – “couple” to a phone – allowing them to “talk”, using sounds, to the system on the other side.
As time went on, these restrictions gradually became more relaxed. Third party devices (with proper certification) could be connected to the analog lines directly, allowing better modulation, data speeds and error correction techniques. But naturally, everything has its limits, and the landline faded into obsolescence.
What about experimenting with the old phone? Well, acoustic couplers were not used extensively in the part of the world I grew up. Apart from a speaker and a mike, the rest can be reconstructed in software nowadays. But a simple device crafted as a weekend project will do it too!
Yeah, a 3D printed contraption might not look as sexy as a vintage 1960s wooden box, doing all the magic…
This acoustic coupler uses the 555 timer to transmit audio, an analog PLL with an Exar XR2211 (Fairchild RC2211A, etc) to decode audio and generate carrier detect, and a Maxim MAX232 to convert voltages to and from RS232 levels, so that it can work with cheap USB-serial adapters as well as old computers alike.
Parts from an old handset were used, too: a small speaker/earpiece for transmitting, and a carbon microphone for receiving. This was chosen because the old Tesla phone already has a carbon mic, and they do not require further amplification to work: one bias resistor and a coupling capacitor is enough.
A modern electret microphone can be used by changing the bias resistor (R4, see below for schematic) to around 10K, and adding a suitable preamp between C8 and pin 2 of the XR2211. In that case, it might be better to use a separate supply, as the charge pump of the MAX232 introduces ripple on the +5V rail, which might mix in.
If required, the input sensitivity of the 2211 can be decreased with a parallel resistor of a few hundred kilohms, connected between pin 2 and ground.
Left: original carbon microphone of the 1950s Tesla Model “K” telephone, with some rust;
right: a later variant (1970s) with a makeshift flat insert. This can sit flush inside the coupler.
The most crucial part is a tight, soundproof fit between the phone and the coupler. The receiving microphone must be positioned very close to the handset earpiece: compared to smartphones, these are faint.
In addition, bushings/gaskets should be used to mount the microphone and insulate the coupler from outside noise. If carbon microphones are involved (handset, coupler or both), the setup will be more sensitive to vibration than with modern mikes. Any outside noise or vibration with an amplitude greater than the signal from the earphone will break the datastream.
Back in the day, homebrew designs used drain pipe adapters, various foam rings and so forth. Was never a one-size-fits-all 🙂
It’s an ad hoc construction, but good enough for a weekend project
The schematic values below are for a Bell 103 compatible modulation, with a baud rate of 300 bps full duplex. The old phone might do with up to 1200 bps with half-duplex and good sound insulation, but I wanted to keep it simple.
It is an analog circuit, so it will require some fine tuning for proper operation, but the initial values of the trimpots should be good enough for a start. P1, P3 and P4 need to be multi turn. P2, the output volume control, is a normal one and is set somewhere in the middle.
The transmitter (NE555-driven earphone) will normally stay quiet, as the DTR (data terminal ready) line is deasserted (Q2 closed). A serial cable needs to be connected and a terminal application opened; this should assert the DTR line and produce an audible tone. The 555 is then adjusted for an idle (“mark”) tone of 1270 Hz with trimpot P2. The transmitting tone (“space”, with transistor Q1 closed during TX), is set to a frequency of 1070 Hz using trimpot P1. Tinkering with the volume control will cause these to drift a bit, but these are not that crucial – and the volume should not be set too loud anyway.
A smartphone or other tone generator app can be set to calibrate these two sound frequencies; oscilloscope or signal generator level precision is not required.
Schematic, with the initial values set for Bell 103 at 300 bps.
For simplicity, the setup is pre-set for “originate mode” (phone always makes outgoing calls). CTS is always asserted for full duplex.
The frequencies can be tuned down for CCITT V.21 operation, if wished.
With the transmitter set up, the XR2211 receiver can then be adjusted by dialing into a modem (with handshaking configured for Bell 103), or by using minimodem –tx -8 -M 2225 -S 2025.
Trimpot P4 is set at initial 20K and slowly adjusted so that the data carrier detect (DCD) LED is on when receiving a “mark” (idle) frequency of 2225 Hz, and that proper data is being decoded in the terminal console, receiving 2025 Hz. This will be more finicky to tune than the transmitter – and, if the coupling is not good enough, or external noise is being picked up, it will not be possible to get right!
To use V.21 modulation instead of Bell 103 at the same data rate, first P2 of the 555 is tuned down to 1180 Hz (idle) and then P1 to 980 Hz (transmit). Afterwards, P4 is set at initial 26K and slowly adjusted until decoding is right: either on a real modem, or with minimodem –tx 8 -M 1850 -S 1650.
Good ole HyperTerminal: modem on the left, acoustic coupler on the right.
The modem (Tornado Netherlands, chipset Davicom DM336P) auto-negotiates down to 1200bps. 300bps needs to be enabled manually with AT*N17 (Bell) or AT*N18 (V21).
Modem DTE speed is at 115200bps, so hardware flow control has to be on for proper operation.
I also made a Windows INF driver definition for the acoustic coupler, utilizing the generic Unimodem driver. That way, the modem will appear as a device that could be used for a dial up connection. There are two issues, though: 300bps is much too slow for a TCP/IP stack, even with all sorts of registry hacks and timeouts increased, and with MTU down from 1500 to 128…
The second issue is that the Unimodem driver expects Hayes text responses to switch from a “command mode” to a “data mode”. There isn’t a way on how to trigger it using the status of the carrier detect signal alone, which this acoustic coupler provides. As such, the INF driver will accept any 1st decoded byte as a command to start interpreting data. However, losing carrier detect will make Windows drop the line.
Winsock is not being patient enough
Debian Linux running pppd is used as a dial-up server