Shared-Medium Access & Resource Division

A shared medium cannot accept arbitrary simultaneous transmissions unless receivers can separate them. A medium-access mechanism therefore partitions a resource, schedules it, or resolves competing attempts. The same physical dimensions—time, frequency, code, and space—appear in both multiplexing (combining streams) and multiple access (coordinating independent transmitters); the operational distinction is who controls the allocation.

One resource, four separation axes #

Let total usable time, bandwidth, and spatial/code discrimination form a resource region $\mathcal R$. Each transmission occupies a subset $\mathcal R_i$. Successful coexistence requires either

$$\mathcal R_i\cap\mathcal R_j=\varnothing$$

in a dimension the receiver observes, or sufficiently low cross-coupling that both meet their signal-to-interference requirements. “Orthogonal” allocations idealize the first case; real filters, timing error, multipath, and sidelobes produce guard costs and leakage.

Time division #

TDMA gives users distinct intervals. If a frame lasts $T_f$ and user $i$ owns payload time $T_i$, its ideal duty factor is $T_i/T_f$. Guard intervals, synchronization, control slots, and unused reservations reduce utilization. Static slots bound access delay but waste capacity when owners are idle; dynamic scheduling reallocates slots but needs requests, grants, and scheduler state.

Frequency division #

FDMA assigns bands; WDM applies the same idea to optical wavelengths. Filters need transition bands, oscillators have error, and adjacent allocations leak, so guard bands and emission masks are part of capacity. OFDMA divides time-frequency resources into many orthogonal subcarriers and time intervals that a scheduler can allocate dynamically.

Code division #

CDMA overlays signals in time and frequency but assigns sequences whose correlations let a receiver separate them under stated alignment and power conditions. Cross-correlation, near–far power imbalance, and multipath determine residual interference. A code name does not create isolation by itself; see spread-spectrum codes.

Spatial division #

Separate wires and fibers are the simplest spatial channels. Antenna arrays, sectorization, and beamforming reuse time-frequency resources when channel responses are distinguishable. Spatial reuse is therefore conditional on geometry, propagation, interference, and receiver processing—not merely physical distance.

Allocation policies #

Centralized allocation can optimize a global objective but depends on timely channel and queue information. Distributed allocation avoids one scheduler but decisions are based on partial, delayed observations. Hidden terminals in wireless are the canonical example: two transmitters can both perceive an idle channel while interfering at a receiver.

Capacity accounting #

Raw signaling rate is not application throughput. For a scheduled resource,

$$R_{useful}=R_{raw}\,\eta_{phy}\,\eta_{guards}\,\eta_{control}\,\eta_{protocol}\,\eta_{retries},$$

where each $0\le\eta\le1$ accounts for coding/preambles, guard time or band, scheduling/control, headers, and failed transmissions. The factors may depend on load: contention retries and scheduler control traffic usually worsen as demand rises.

Fairness also needs a declared unit. Equal airtime is not equal bytes when users have different physical rates; equal throughput can make slow users consume most airtime; equal resource blocks may yield different delivered rates under different channel quality.

Ethernet and Wi-Fi are no longer one collision story #

Early shared coaxial and hub Ethernet used CSMA/CD because multiple stations transmitted into one collision domain. A modern full-duplex point-to-point Ethernet link has one transmitter in each direction and no shared-medium contention on that link; switches arbitrate output queues internally instead.

IEEE 802.11 wireless LANs still coordinate access to shared radio resources. Their MAC includes distributed contention and scheduled/coordinated mechanisms because a station cannot assume its local observation describes interference at every receiver. Treating Wi-Fi as “wireless Ethernet” hides the access-control problem that dominates latency under load.

Diagnose access problems #

• Establish whether the link is point-to-point full duplex, shared wired, or shared radio before discussing collisions.
• Measure airtime/resource occupancy, not only delivered bytes.
• Separate scheduled wait, contention wait, retransmission, and queueing after access.
• Identify guard, preamble, control, acknowledgement, and retry overhead at the observed rate.
• Test fairness in the intended unit and include low-rate or distant participants.
• Verify whether supposedly independent allocations leak through filters, correlation, beams, or a shared lower bottleneck.

Medium access is correct only when every transmitter has a rule for obtaining the resource, every receiver can separate the allowed simultaneous signals, and overload behavior is explicit.