As early experiments in sending voice packets over the ARPANET revealed, voice transmission required a pure datagram service. The ARPANET with its “Ready for Next Message” (RFNM) signaling technique was all about reliable end-to-end deliver of information. New packets could not be sent into the network until outstanding packets that had already been sent were acknowledged as having successfully gotten to their destination host computer by a RFNM message. This did not work for voice. Voice packets needed to be sent without waiting for an end-to-end acknowledgment. If a small number of packets were lost along the way or arrived out of order and were discarded by the receiver, the human ear and brain could interpolate through the missing audio information and understand what was being said. If a large number of packets were lost, listening to the voice could become tiresome and possibly unintelligible. When users were trying to have a two-way conversation, they were also sensitive to the end-to-end delay of the voice packets. Experiments showed that when end-to-end delay rose above 200 msec., users felt that they were participating in a “half-duplex” conversation in which people had to be very deliberate in turn-taking and not interrupt each other. It became obvious to the early packet voice researchers that the underlying packet switched network would have to deliver a certain level of service ( low packet loss, low end-to-end delay, and low inter-packet jitter - the time-difference of arrival of voice packets at the receiver) for users to feel that packet voice was an acceptable substitute for the circuit switched voice that they were used to with the telephone system.
ARPA IPTO realized that it would be necessary to create a packet switched network that could deliver the right level of quality of service (QoS) to carry voice along with data traffic. Even with compression algorithms such as LPC, to get a meaningful large number of packet voice calls to inter-mix with packet data, would require a much higher bandwidth network. High-speed terrestrial circuits such as T1 phone lines that ran at 1.5Mbps were still very expensive and difficult to obtain in the late 1970s. Dick Binder of BBN, Estil Hoversten and Irwin Jacobs of Linkabit Corp., and Bob Kahn and Vint Cerf from ARPA conceived the Wideband Packet Satellite Network (Wideband Net) as an appropriate vehicle to deliver the necessary QoS. This network was built around a 3Mbps broadcast satellite channel that connected multiple sites in the US and supported broadcast and multicast delivery for voice conferencing. The Wideband Net added packet speech to the packet satellite data networking experiments by creating a stream service for packet voice traffic. The stream service permitted sites to reserve periodic time slots in each frame on the satellite channel to carry the voice packets. The rest of the frame could be used for more bursty data traffic. The first four sites were the MIT Lincoln Lab in Lexington, MA, the Defense Communications Engineering Center (DCEC) in Reston, VA, USC ISI in Marina del Rey, CA, and the Stanford Research Institute (SRI) in Menlo Park, CA.
Binder led the BBN team that built the original Wideband Net packet switch; Gil Falk took over the project when Binder left BBN. The switch was built on the BBN Pluribus multiprocessor and was called the Pluribus Satellite IMP or PSAT for short. The Pluribus was chosen because with about a half dozen Lockheed SUE minicomputer processors and a high-speed satellite interface, it had the processing power to run the PODA algorithms and keep up with the 3Mb/s channel. The Pluribus presented a very difficult programming environment. During the project’s early stages many people including John Robinson, Tony Lake, Jane Barnett, Dick Koolish, Steve Groff, Walter Milliken, Marian Nodine, and Steve Blumenthal worked on the implementation. Burnout on the programming team was a problem. The PSAT software was buggy and could not be made to run reliably for any lengthy period of time. The hardware’s several wire-wrapped boards also had some long term stability problems and faults were difficult to isolate to specific hardware or software causes. Blumenthal took on an operations role and began to figure out how to make the PSAT and the entire system more robust overall, and eventually became Wideband Net project manager.