Aggressive Non-Orthogonality
ACM HotNets ’26 · Salt Lake City

What if more intentional interference improved connectivity?

In device-to-satellite links, massive IoT and emergency or tactical networks, signals are often extremely weak and many devices must share limited wireless resources. At low SNR, letting many weak users transmit simultaneously, even to a single receive antenna, can substantially increase capacity and connectivity. We call this aggressive non-orthogonality.

4
The received superimposed constellation at a single-antenna receiver, with BPSK users transmitting concurrently, each arriving at 4 dB SNR. Every dot is one received sample. With more users, each received symbol is the superposition of several users’ symbols, making it progressively more challenging to jointly decode the information streams.
The paper · HotNets ’26 Enabling “Aggressive” Non-Orthogonal Transmissions for Improved Connectivity in the Low SNR Regime Chathura Jayawardena, George N. Katsaros, Konstantinos Nikitopoulos PDF · doi.org/10.1145/3847290.3859970

Revisiting information theory

Traditionally, wireless systems avoid interference by separating users in time, frequency or space. At low SNR, however, the bottleneck is noise rather than interference: each device is power-limited, so the only way to collect more received energy is to let more of them transmit simultaneously. Capacity then grows almost linearly with that pooled energy. The users don’t coordinate and don’t send the same signal; each adds its own channel gain to the total, whatever the others send.

Adjust the number of users below and compare with orthogonal multiple access (OMA, e.g., TDMA). Then increase the SNR and watch the advantage shrink: this is a low-SNR effect.

Concurrently transmitting users–bits/s/Hz
Orthogonal multiple access–bits/s/Hz
Gain vs OMA–sum capacity ratio
Concurrently transmitting users (non-orthogonal) Orthogonal multiple access
4
0 dB
Receive antennas

Ergodic sum capacity, log₂det(I + (Pₜ/σ²)HHᴴ), averaged over 1,500 independent Rayleigh channel draws.

In the paper: §2, Eq. (2)–(3), Fig. 1. Here channels are i.i.d. Rayleigh; the paper uses the 3GPP TDL-A model with medium spatial correlation, so absolute values differ.

Making it practical

However, realizing these theoretical gains requires rethinking conventional receiver architectures: revisiting the fundamentals of optimal soft-output detection and exploiting the structure of channel codes to enable practical maximum a posteriori (MAP) processing.

We built it on software-defined radios. Four single-antenna users concurrently transmit to one single-antenna receiver, which detects and decodes them over the air in real time (demo setup below). To compare schemes fairly, captured channels are then replayed offline so every scheme sees identical conditions: with four users, our receiver reaches more than twice the spectral efficiency of orthogonal multiple access.

Testbed: five NI USRP X310 software-defined radios (four single-antenna transmitters, one single-antenna receiver). 3.4 GHz carrier, 40 MS/s, OFDM with a 128-point FFT and a 32-sample cyclic prefix.

Sum spectral efficiency with a single receive antenna

Average SNR

Setup: captured channels replayed offline, so every scheme sees identical conditions. One receive antenna throughout.

How many devices fit at a fixed quality of service?

Orthogonal multiple access Linear MMSE Proposed joint soft detection Proposed joint MAP detection-decoding
Service per device

The aggregate rate is shared, so lighter services fit more devices: up to 10× as many devices as orthogonal multiple access with a single receive antenna.

Setup: 4 dB SNR, 3GPP TDL-A channel, 30 kHz subcarrier spacing. Broadband users get three resource blocks (1.08 MHz) each; voice and text use the minimum allocation of one resource block (360 kHz).

In the paper: §4, Fig. 2 spectral efficiency, §5, Fig. 4 supported devices.