Ferkans — Interactive Telecom Tutor

Degrees of Freedom: The High-SNR Capacity Slope

When the exact capacity of a multiuser channel is unknown (as for the IC), the degrees of freedom (DoF) framework provides a powerful asymptotic characterisation. The DoF captures the pre-log factor — how many independent signal dimensions the network supports at high SNR. For the single-user MIMO channel, the DoF equals $\min(n_t, n_r)$ . For the $K$ -user IC, the surprising answer is $K/2$ — each user gets half a degree of freedom, regardless of $K$ , achieved through interference alignment. This result, due to Cadambe and Jafar (2008), fundamentally changed our understanding of wireless networks by showing that interference does not limit the scaling of multiuser capacity.

Definition:
Degrees of Freedom

The degrees of freedom (DoF) of a channel is defined as the capacity pre-log factor:

$d = \lim_{\text{SNR} \to \infty} \frac{C(\text{SNR})}{\frac{1}{2}\log_2(\text{SNR})}$

For a multiuser channel, the DoF region is the set of all DoF tuples $(d_1, \ldots, d_K)$ where:

$d_k = \lim_{\text{SNR} \to \infty} \frac{R_k(\text{SNR})}{\frac{1}{2}\log_2(\text{SNR})}$

for some achievable rate tuple $(R_1, \ldots, R_K)$ in the capacity region.

The sum DoF is: $d_{\Sigma} = \max \sum_{k=1}^K d_k$

Examples:

Single-user SISO: $d = 1$
Single-user MIMO ( $n_t \times n_r$ ): $d = \min(n_t, n_r)$
$K$ -user SISO MAC: $d_{\Sigma} = 1$ (bottleneck at the single receive antenna)
$K$ -user SISO BC: $d_{\Sigma} = 1$ (same reason)
$K$ -user SISO IC: $d_{\Sigma} = K/2$ (Cadambe-Jafar)

DoF analysis ignores the constant gap between the achievable rate and $d \cdot \frac{1}{2}\log_2(\text{SNR})$ . It is most informative at high SNR and for understanding scaling behaviour, but may not reflect performance at practical SNR levels.

,

Theorem: DoF of the $K$ -User Interference Channel

The $K$ -user time-varying/frequency-selective SISO interference channel with i.i.d. channel coefficients drawn from a continuous distribution has sum DoF:

$d_{\Sigma} = \frac{K}{2}$

with $d_k = 1/2$ per user, achieved by interference alignment (IA).

At first glance, the IC with $K$ users seems limited to $d_{\Sigma} = 1$ (TDMA) because each receiver has a single antenna and sees $K-1$ interferers. IA shows that by carefully designing transmit signals using channel variations (time/frequency), all interfering signals at each receiver can be aligned into a subspace occupying half the signal dimensions, leaving the other half free for the desired signal. Each user thus achieves $d = 1/2$ , and the total is $K/2$ — unbounded as $K$ grows.

Proof

Achievability via interference alignment

Consider the $K$ -user IC over $n$ time/frequency slots. Channel from transmitter $j$ to receiver $k$ in slot $t$ : $h_{kj}(t)$ . Stack into diagonal matrices $\mathbf{H}_{kj} = \mathrm{diag}(h_{kj}(1), \ldots, h_{kj}(n))$ .

Each transmitter $k$ uses a precoding matrix $\mathbf{V}_k \in \mathbb{C}^{n \times d}$ ( $d$ streams).

Alignment conditions: For each receiver $k$ : $\mathrm{span}(\mathbf{H}_{kj}\mathbf{V}_j) \subseteq \mathbf{S}_k, \quad \forall j \neq k$

where $\mathbf{S}_k$ is the interference subspace at receiver $k$ with $\dim(\mathbf{S}_k) \leq n - d$ .

Desired signal condition: $\mathrm{rank}(\mathbf{P}_k^{\perp} \mathbf{H}_{kk}\mathbf{V}_k) = d$

where $\mathbf{P}_k^{\perp}$ projects onto the complement of $\mathbf{S}_k$ .

Feasibility for $K = 3$

For $K = 3$ users, $d = n/2$ streams each (total $3n/2$ DoF over $n$ dimensions, normalised to $3/2$ DoF per dimension).

Choose $n = (M+1)^3$ for large $M$ and set: $\mathbf{V}_1 = \mathrm{span}\{(\mathbf{H}_{12}^{-1}\mathbf{H}_{13})^a (\mathbf{H}_{12}^{-1}\mathbf{H}_{32}^{-1}\mathbf{H}_{31})^b \mathbf{e} : 0 \leq a+b \leq M\}$

for some generic vector $\mathbf{e}$ , with similar constructions for $\mathbf{V}_2, \mathbf{V}_3$ .

The key algebraic property: at each receiver, interference from two undesired transmitters aligns into the same subspace. Specifically, at receiver 1: $\mathrm{span}(\mathbf{H}_{12}\mathbf{V}_2) = \mathrm{span}(\mathbf{H}_{13}\mathbf{V}_3)$

This alignment ensures the interference subspace has dimension $\leq n/2 + o(n)$ , leaving $n/2$ dimensions for the desired signal.

Converse (upper bound)

The upper bound $d_{\Sigma} \leq K/2$ follows from considering any user pair $(k, j)$ . Providing all other users' messages as side information reduces the IC to a two-user channel with DoF $\leq 1$ per pair.

More precisely, for each user $k$ : $d_k \leq 1$

and for each pair $(k, j)$ : $d_k + d_j \leq 1$

Summing all $\binom{K}{2}$ pairwise bounds and using symmetry: $(K-1) d_{\Sigma} \leq \binom{K}{2}$ , giving $d_{\Sigma} \leq K/2$ . $\blacksquare$

Definition:
Interference Alignment

Interference alignment (IA) is a linear precoding strategy where each transmitter designs its beamforming vectors so that at each receiver, all interfering signals are confined to a low-dimensional subspace, leaving the remaining dimensions free for the desired signal.

Formal condition at receiver $k$ : $\bigoplus_{j \neq k} \mathrm{span}(\mathbf{H}_{kj}\mathbf{V}_j) = \mathbf{S}_k$ $\dim(\mathbf{S}_k) + d_k \leq n$ $\mathbf{H}_{kk}\mathbf{V}_k \cap \mathbf{S}_k = \{\mathbf{0}\}$

Practical challenges:

Requires global CSI at all transmitters.
The required signal space dimension $n$ grows exponentially in $K$ for exact alignment.
At finite SNR, IA may be suboptimal because it ignores noise enhancement.

IA is often described with the analogy: imagine $K$ interfering transmissions in a room. IA arranges them so that from each receiver's perspective, all interference casts "shadows" that overlap in the same direction, leaving a clear view of the desired signal in the orthogonal direction.

Interference Alignment in 2D Signal Space

Visualise how interference alignment works for the 3-user IC over 2 symbol extensions. At receiver 1, interference from users 2 and 3 initially points in different directions. The precoding vectors are designed so that both interference signals align into a single 1D subspace, leaving the orthogonal dimension free for the desired signal from user 1.

At each receiver, all

K-1

interference signals align into a subspace of dimension

n/2

, leaving

n/2

dimensions for the desired signal. Total DoF:

K/2

.

DoF Scaling with Number of Users

Compare the sum DoF of the $K$ -user interference channel (achieved by interference alignment) with TDMA and the MIMO BC. The IA DoF scales as $K/2$ , while TDMA is fixed at 1 regardless of $K$ . The MIMO BC DoF is limited by $\min(n_t, K)$ . Adjust the maximum number of users $K$ and the number of transmit antennas $n_t$ (for MIMO BC comparison) to visualise the scaling.

Parameters

Maximum number of users

K

20

Transmit antennas

n_t

(MIMO BC comparison)1

Example: Interference Alignment for 3 Users

Consider the 3-user SISO IC over $n = 2$ symbol extensions (time-varying channels).

(a) What is the target DoF per user? (b) Design alignment vectors for $d = 1$ stream per user. (c) Verify the alignment conditions at each receiver. (d) Why does $n = 2$ work for $K = 3$ but not $K = 4$ ?

Solution

Target DoF

(a) Total DoF $= K/2 = 3/2$ . Per user: $d = 1/2$ . Over $n = 2$ extensions: each user sends $d = 1$ stream (1 out of 2 dimensions), achieving $d = 1/2$ per c.u.

Alignment design

(b) At receiver 1, interference from users 2 and 3 must align: $\mathrm{span}(\mathbf{H}_{12}\mathbf{v}_{2}) = \mathrm{span}(\mathbf{H}_{13}\mathbf{v}_{3})$

Choose $\mathbf{v}_{2} = [1, 1]^T$ and $\mathbf{v}_{3} = \mathbf{H}_{13}^{-1}\mathbf{H}_{12}\mathbf{v}_{2}$ .

Similarly at receivers 2 and 3, choose $\mathbf{v}_{1}$ to align interference. For generic (i.i.d. continuous) channels, a consistent solution exists.

Verification

(c) At receiver 1: interference spans $\mathrm{span}(\mathbf{H}_{12}\mathbf{v}_{2}) = \mathrm{span}(\mathbf{H}_{13}\mathbf{v}_{3})$ — a 1D subspace.

Desired signal: $\mathbf{H}_{11}\mathbf{v}_{1}$ — a 1D subspace.

For generic channels, $\mathbf{H}_{11}\mathbf{v}_{1}$ is linearly independent of the interference subspace (almost surely). Receiver 1 can zero-force the interference and decode.

Why not $K = 4$

(d) For $K = 4$ , each receiver sees 3 interferers. With $n = 2$ dimensions and $d = 1$ stream per user, the interference occupies 3 dimensions — but we only have 2. We need $n$ large enough that $d(K-1) \leq n - d$ , i.e., $n \geq dK$ . For $K = 4$ , $d = 1$ : need $n \geq 4$ . The symbol extension grows with $K$ . $\blacksquare$

Quick Check

What is the sum DoF of the 10-user SISO interference channel with time-varying channels?

1, because there is only one receive antenna per user

5, achieved by interference alignment

10, because each user gets 1 DoF

2, because it is limited by the pairwise constraint $d_k + d_j \leq 1$

Correction:

5, achieved by interference alignment

The Cadambe-Jafar result shows that the $K$ -user SISO IC with time-varying channels has sum DoF $= K/2$ . For $K = 10$ , this gives $d_{\Sigma} = 5$ , with each user achieving $d_k = 1/2$ . This is achieved by interference alignment over sufficiently long symbol extensions. The pairwise constraint $d_k + d_j \leq 1$ gives the upper bound $d_{\Sigma} \leq K/2$ , and IA achieves it.

Why This Matters: Interference Alignment in the Delay-Doppler Domain

The interference alignment framework assumes time-varying or frequency-selective channels to create the signal-space diversity needed for alignment. In the delay-Doppler domain used by OTFS modulation, the channel is naturally sparse and structured, providing a different (and potentially more efficient) basis for alignment. The OTFS book explores how delay-Doppler domain precoding relates to interference alignment and how OTFS waveforms can be designed to exploit the channel structure for interference management in high-mobility multiuser scenarios.

Degrees of Freedom (DoF)

The high-SNR capacity pre-log factor: $d = \lim_{\text{SNR}\to\infty} C(\text{SNR})/\frac{1}{2}\log_2(\text{SNR})$ . Captures how many independent signal dimensions the channel supports.

Related: Interference Alignment (IA)

Interference Alignment (IA)

A precoding strategy where transmitters coordinate their beamforming so that at each receiver, all interference aligns into a subspace of minimum dimension. Achieves the optimal DoF of $K/2$ for the $K$ -user IC.

Related: Degrees of Freedom (DoF)

Degrees of Freedom Framework