Time-Division Multiplexing: How Deterministic Networks Work

On Friday, 13 November 2026, the US Federal Aviation Administration (FAA) issued a rare nationwide ground stop that halted departing flights across the country for more than four hours, delaying or cancelling nearly 7,000 services. In the press briefing that followed, US Transportation Secretary Sean Duffy pointed directly at the culprit: overloaded switches and routers in the FAA’s ageing Time-Division Multiplexing (TDM) telecoms network began producing data inconsistencies, forcing officials to pause departures while newer IP-based switches stayed online without issue. For most travellers, “TDM” sounded like an obscure piece of bureaucratic jargon. For network engineers, it was a sharp reminder of how much critical infrastructure still runs on circuit-switched telecoms technology designed in the mid-twentieth century. Here is how TDM actually works on the wire, why aviation relied on it for so long, and why migrating off it involves far more than simply swapping out a few routers.

TDM

Suppose you need to connect an airport control tower to a radar site ten miles away. Inside the tower, thirty separate systems need to talk to that remote site at the exact same time: controller radio channels, radar plot feeds, weather station telemetry, and direct emergency lines.

Laying thirty separate copper cables across ten miles of countryside is expensive and awkward to maintain. Instead, telecommunications engineers solved this problem decades ago by putting all thirty signals onto a single pair of wires. They did it not by mixing the signals together, but by giving each signal a strict, recurring turn on the wire.

This technique is called Time-Division Multiplexing (TDM). Although telecoms operators across the UK and Europe are actively switching off their legacy TDM networks in favour of IP, the technology still sits underneath plenty of critical infrastructure, from air traffic control centres to electrical substations and railway signalling lines. Understanding how TDM works at the wire level also explains why replacing it with modern Ethernet and IP has proven far trickier than many expected.

The Basic Idea: Taking Turns at High Speed

Imagine a single-lane bridge with a traffic controller at each end. If thirty cars need to cross, the controller lets one car from Lane 1 pass, then one from Lane 2, and so on up to Lane 30, before starting back at Lane 1.

In a TDM circuit, the cars are small chunks of digital data, and the traffic controller is an electronic switch called a multiplexer. At the far end of the cable, a demultiplexer performs the exact reverse operation, routing each incoming chunk to its corresponding output port.

Because this rotation happens thousands of times per second, the gaps between turns are imperceptible. A pilot speaking over a radio link experiences continuous, unbroken audio, even though their voice is actually being chopped into tiny slices and interleaved with twenty-nine other data streams.

Turning Voice into Bytes: PCM and the 125-Microsecond Clock

Before you can interleave multiple signals in time, you have to digitise them. For voice circuits, this relies on Pulse-Code Modulation (PCM), invented by the British engineer Alec Reeves in 1937.

Human speech carries enough intelligibility for telephony within a frequency band of 300 Hz to 3.4 kHz. According to the Nyquist-Shannon sampling theorem, capturing a signal without aliasing requires sampling at more than twice its highest frequency. Telecoms standards settled on an 8 kHz sampling rate, meaning the analogue waveform is measured once every 125 microseconds (1 / 8,000).

Each voltage measurement is quantised into an 8-bit integer. Rather than spacing the quantisation steps linearly, PCM uses a logarithmic curve, so that quiet speech gets finer resolution than loud bursts. Multiplying 8,000 samples per second by 8 bits per sample yields a constant bit rate of 64 kbps. In telecoms terminology, this 64 kbps channel is the fundamental building block of digital voice networks, known as a DS0 or E0 circuit.

Framing, E1, and T1

Once each channel is reduced to a steady stream of 8-bit bytes arriving every 125 microseconds, the multiplexer packs them into repeating frames. In the UK and across Europe, the standard building block is the E1 circuit, defined by ITU-T recommendations G.703 and G.704.

An E1 frame contains 32 time slots, numbered TS0 through TS31, with each slot holding 8 bits:

• TS0 (Framing and Alignment): Carries the Frame Alignment Signal (FAS), a known bit pattern that allows the receiving demultiplexer to lock onto the start of each frame, along with CRC-4 bits for error detection.

• TS1 to TS15 and TS17 to TS31 (Payload): Thirty independent 64 kbps bearer channels for voice or data.

• TS16 (Signalling): Reserved for call setup, teardown, and line state information, using either Channel Associated Signalling (CAS) or Common Channel Signalling (CCS) such as ISDN PRI and SS7.

Adding those 32 slots together gives 256 bits per frame. Transmitted 8,000 times per second, an E1 link runs at an exact line rate of 2.048 Mbps.

North America and Japan adopted a slightly different standard called T1 (or DS1). A T1 frame bundles 24 channels of 8 bits each (192 bits) plus a single framing bit at the front, making 193 bits every 125 microseconds, or 1.544 Mbps. Rather than dedicating a full time slot to signalling, T1 traditionally used robbed-bit signalling, stealing the least significant bit from every sixth frame to pass line states.

The Hidden Complexity of Clocking

Look closely at an E1 frame and you will notice something missing compared to an Ethernet frame or an IP packet: there are no destination addresses, no port numbers, and no sequence headers.

In TDM, identity is purely positional. Channel 4 belongs to the approach radar simply because it arrives in TS4, exactly 31.25 microseconds after the start of TS0. Because there are no packet headers to guide the hardware, both ends of the link must run on tightly synchronised clocks.

If the transmitter’s clock runs slightly faster than the receiver’s clock, bits will eventually pile up in the receiver’s elastic buffer until it overflows. That forces a frame slip, dropping or duplicating an entire 125-microsecond frame. On a voice call, a single slip sounds like a faint click. On a radar data feed or a high-speed cryptographic link, it causes immediate framing loss and resynchronisation delays.

To prevent drift, national telecoms networks maintain a strict timing hierarchy:

1. Stratum 1 (Primary Reference Clock): Caesium atomic clocks or GNSS-disciplined oscillators accurate to 1 \times 10^{-11}, which works out to roughly one frame slip every 72 days if two independent clocks drift at their maximum tolerance.

2. Stratum 2 and 3: Rubidium or high-stability oven-controlled crystal oscillators (OCXOs) deployed in regional exchanges, capable of holding accurate timing for days if the upstream reference link fails.

3. Loop Timing: Customer premises equipment, such as an airport’s multiplexers, typically recovers its clock directly from the incoming bit transitions on the telecoms line itself.

When operators bundled multiple E1s into higher-capacity trunks (such as E2 at 8.448 Mbps and E3 at 34.368 Mbps) under the older Plesiochronous Digital Hierarchy (PDH), slight clock variations between independent E1 sources forced multiplexers to insert stuffing bits to pad out the difference. Extracting a single 64 kbps channel out of a 34 Mbps PDH stream meant demultiplexing the entire hierarchy step by step. That limitation eventually led to SDH (Synchronous Digital Hierarchy) and SONET over optical fibre, which introduced synchronous payload pointers so individual channels could be dropped or inserted directly.

Why Critical Infrastructure Still Cares

From a modern software engineering perspective, TDM looks terribly inefficient. If nobody is speaking on Channel 4, TS4 still travels across the wire filled with idle bit patterns. You cannot borrow unused bandwidth from quiet channels to speed up a busy file transfer.

Yet that exact rigidity is why aviation, rail, and power utilities relied on TDM for so long. TDM provides hard determinism:

• Bounded, constant latency: Propagation delay is fixed by the speed of light in copper or glass plus a tiny, predictable buffer delay in the multiplexers. There is zero queueing jitter.

• Complete isolation: A sudden burst of traffic on Channel 12 cannot starve Channel 1 of bandwidth or cause packet drops. Slot 1 belongs exclusively to Channel 1.

• Simple failure modes: Hardware either holds frame sync or raises an immediate Loss of Signal (LOS) or Alarm Indication Signal (AIS) within milliseconds.

When carriers retire PSTN and TDM leased lines, migrating these workloads onto IP and MPLS networks requires engineering determinism back into a protocol suite built around statistical multiplexing and best-effort delivery. Engineers have to deploy Circuit Emulation Services (such as SAToP or CESoPSN) to wrap raw E1 bitstreams inside UDP/IP packets, pair them with IEEE 1588v2 Precision Time Protocol (PTP) to distribute nanosecond-level clock synchronisation over packet switches, and enforce strict QoS queues so a routine software update never delays a controller’s radio transmission by fifty milliseconds.

TDM may be a twentieth-century architecture, but its core lesson remains relevant whenever we build real-time systems: if you want absolute predictability, you have to reserve time itself.