Why Multiple Access Matters
Every generation of cellular technology has been defined by its multiple access scheme — the method by which a base station divides its radio resources among many users simultaneously. 2G used TDMA/FDMA. 3G brought CDMA. 4G LTE standardised OFDMA. And 5G, while OFDM-based, introduces Non-Orthogonal Multiple Access (NOMA) as a key technique for massive connectivity and improved spectral efficiency.
But what exactly is wrong with the old orthogonal approach? And how does NOMA fix it?
Key insight: In Orthogonal Multiple Access (OMA), the base station can only serve one user per resource block (time slot + sub-carrier). NOMA serves two or more users on the same resource block by using power levels to separate them.
OMA vs NOMA: The Core Difference
In OMA (e.g., OFDMA), User A gets sub-carrier 1 and User B gets sub-carrier 2. They never interfere because they live in different parts of the spectrum. This is clean and simple, but the cost is that if User B has nothing to transmit, that sub-carrier is wasted.
NOMA takes a different approach: both User A and User B transmit on sub-carrier 1, but at different power levels. The base station (in the downlink) deliberately superposes their signals. A nearby, strong user (User B) receives a low-power signal, while a far, weak user (User A) receives a high-power signal.
The composite transmitted signal looks like:
Where $\alpha_1 + \alpha_2 = 1$ are the power allocation coefficients and $P$ is total transmit power.
The SIC Receiver — How to Decode Two Signals in One
The magic happens at the receiver. The strong user (good channel, small path loss) uses a technique called Successive Interference Cancellation (SIC) to decode both messages:
- First, decode User A's signal (the high-power one) by treating User B's signal as noise.
- Reconstruct User A's contribution and subtract it from the received signal.
- Now decode the cleaned-up User B signal.
The weak user (poor channel, high path loss) simply decodes the high-power signal directly — it doesn't need SIC because its own signal dominates.
import numpy as np
def noma_sic(received, h1, h2, alpha1, alpha2, noise_power):
"""
Simulates downlink NOMA SIC decoding.
h1: channel gain of far user (weaker)
h2: channel gain of near user (stronger)
"""
# Step 1: Near user decodes far user's signal first
sinr_step1 = (alpha1 * abs(h2)**2) / (alpha2 * abs(h2)**2 + noise_power)
# Step 2: After SIC, near user decodes own signal
sinr_step2 = (alpha2 * abs(h2)**2) / noise_power
# Far user decodes directly (no SIC needed)
sinr_far = (alpha1 * abs(h1)**2) / (alpha2 * abs(h1)**2 + noise_power)
return sinr_step1, sinr_step2, sinr_far
# Example: far user at 0.1 channel gain, near user at 0.8
h_far = 0.1 + 0j
h_near = 0.8 + 0j
alpha1, alpha2 = 0.8, 0.2 # more power to far user
noise = 0.01
s1, s2, sf = noma_sic(None, h_far, h_near, alpha1, alpha2, noise)
print(f"Near user SINR (step 1): {10*np.log10(s1):.2f} dB")
print(f"Near user SINR (step 2): {10*np.log10(s2):.2f} dB")
print(f"Far user SINR: {10*np.log10(sf):.2f} dB")Rayleigh Fading and Its Impact on NOMA Capacity
In practice, wireless channels aren't static — they fluctuate due to multipath propagation. A Rayleigh fading channel models the case where there is no dominant line-of-sight component between transmitter and receiver, and the channel gain $h$ follows a complex Gaussian distribution:
$\Omega = \mathbb{E}[|h|^2]$ is the average channel power (path loss).
The instantaneous channel capacity for a NOMA user under Rayleigh fading is a random variable. We care about the ergodic capacity (average over many channel realizations):
Because $|h|^2$ is exponentially distributed (for Rayleigh fading), this expectation has a closed-form expression involving the exponential integral $E_1(\cdot)$, which researchers use to compare NOMA vs OMA analytically.
Pitfall: SIC relies on the strong user having a significantly better channel than the weak user. If channel gains are similar, the "strong" user struggles to decode the weak user's signal first, which breaks SIC. Proper power allocation is critical.
Power Allocation: The Key Design Parameter
The power allocation coefficients $\alpha_1$ (far/weak user) and $\alpha_2$ (near/strong user) must be chosen carefully:
- $\alpha_1 > \alpha_2$ always — more power goes to the weaker user, otherwise SIC fails.
- A typical split is 80/20 or 70/30, but optimal values depend on the channel conditions and target QoS.
- Dynamic power allocation based on real-time CSI (Channel State Information) feedback is an active research area.
Spectral Efficiency Comparison
The sum rate of a 2-user NOMA system is always greater than or equal to OMA, and strictly greater when the two users have different channel qualities. This can be proven via the rate-splitting argument — NOMA achieves a rate region that strictly contains the OMA rate region.
Bottom line: NOMA is particularly powerful in heterogeneous scenarios — cell-edge vs. cell-center users — which are exactly the conditions that dominate real 5G deployments. It's why 3GPP has included NOMA variants in the 5G NR specification discussions.
Conclusion
NOMA represents a paradigm shift in how we think about radio resource management. By embracing interference rather than eliminating it, and using smart cancellation at the receiver, it squeezes more capacity out of the same spectrum. The trade-off is receiver complexity and sensitivity to power allocation errors — but these are engineering problems that 5G chipsets are increasingly solving in hardware.
In upcoming posts, we'll extend this analysis to multi-user NOMA clusters and explore how machine learning can drive dynamic power allocation in real-time.