Ferkans — Interactive Telecom Tutor

Pilot Assignment as a Design Variable

Given that pilot contamination cannot be entirely avoided in multi-cell systems, a natural question is: can the assignment of pilot sequences to users be optimized to minimize contamination?

The severity of pilot contamination between two users depends critically on their spatial covariance matrices $\mathbf{R}_{1,k}$ and $\mathbf{R}_{\ell,k}$ . If the covariance matrices are nearly orthogonal (users are in very different angular directions), the SINR floor is high. If they are nearly collinear (users in similar directions from their respective base stations), the floor is low.

Pilot assignment asks: given a set of users with known covariance matrices, which users should share pilots to maximize system-level performance?

Definition:
Pilot Contamination Metric

For users $k$ in cell 1 and $k'$ in cell $\ell$ , define the inter-cell pilot contamination coefficient:

$\rho_{k,k'} = \frac{\|\mathbf{R}_{1,k}^{1/2}\mathbf{R}_{\ell,k'}^{1/2}\|_F^2} {\|\mathbf{R}_{1,k}\|_F \|\mathbf{R}_{\ell,k'}\|_F} \in [0, 1]$

This measures the "overlap" between the two users' spatial covariance matrices. $\rho_{k,k'} = 0$ means orthogonal covariance subspaces (no contamination impact). $\rho_{k,k'} = 1$ means identical covariance structure (maximum contamination).

For the one-ring channel model, $\rho_{k,k'}$ is large when users have similar angles of arrival at their respective base stations (similar angular windows).

Greedy Pilot Assignment Algorithm

Complexity:

\mathcal{O}(K_{\text{tot}}^2 \tau_p)

Input: Users

\{1,\ldots,K_{\text{tot}}\}

, pilot pool size

\tau_p

,

contamination metric

\rho_{k,k'}

for all pairs

Output: Pilot assignment

\phi: \{1,\ldots,K_{\text{tot}}\} \to \{1,\ldots,\tau_p\}

1. Sort users by channel strength (largest

\text{tr}(\mathbf{R}_k)

first)

2. Assign pilot 1 to first user (arbitrary starting point)

3. for each remaining user

k

in sorted order do

4. for each pilot

p = 1,\ldots,\tau_p

do

5. Compute cost

C(k, p) = \sum_{j: \phi(j)=p} \rho_{k,j}

// contamination if

k

uses pilot

p

6. end for

7. Assign

\phi(k) = \arg\min_p C(k, p)

// Least contaminating pilot

8. end for

9. return

\phi

Greedy assignment is suboptimal in general but runs in polynomial time and gives good practical results. The optimal assignment minimizing total contamination is an NP-hard combinatorial optimization (related to graph coloring).

Graph-Coloring Pilot Assignment

Complexity:

\mathcal{O}(K_{\text{tot}}^2)

for graph construction,

\mathcal{O}(K_{\text{tot}} + |E|)

for coloring

Input: Users

\{1,\ldots,K_{\text{tot}}\}

, contamination threshold

\rho_{\min}

,

pilot pool size

\tau_p

Output: Pilot assignment

\phi

Step 1: Build contamination graph $G = (V, E)$

1. Nodes

V

: all users across all cells

2. Edge

(k, k')

if

\rho_{k,k'} > \rho_{\min}

// high contamination pair

Step 2: Color graph with $\tau_p$ colors

3. Apply greedy graph-coloring with

\tau_p

colors:

4. for each node

v

in decreasing degree order do

5. Assign smallest color not used by any neighbor of

v

6. end for

Step 3: Assign pilots

7. Map color

c(k)

to pilot sequence

\phi(k) = c(k)

8. return

\phi

Note: If graph needs more than

\tau_p

colors, increase

\tau_p

or relax

\rho_{\min}

The graph-coloring interpretation makes pilot assignment equivalent to proper graph coloring: users connected by high-contamination edges should receive different colors (pilot sequences). The chromatic number of the graph gives the minimum pilot pool size needed to avoid severe contamination.

Example: Pilot Assignment in a Three-Cell System

Three cells each have $K = 2$ users. The contamination metrics are: $\rho_{1,3} = 0.9$ (user 1 cell 1 and user 3 cell 2 are highly contaminated), $\rho_{2,4} = 0.8$ (user 2 cell 1 and user 4 cell 2 are moderately contaminated), $\rho_{1,5} = 0.1$ , $\rho_{2,6} = 0.15$ (cell 3 causes little contamination). Available pilots: $\tau_p = 2$ . Find the greedy assignment.

Solution

Build the contamination graph

Edges (with $\rho_{\min} = 0.3$ ): $(1,3)$ , $(2,4)$ . Users $\{5,6\}$ are isolated.

Apply greedy coloring

User 1: assign pilot 1. User 3 (connected to 1): assign pilot 2.
User 2: assign pilot 1. User 4 (connected to 2): assign pilot 2.
User 5: assign pilot 1. User 6: assign pilot 1 (no conflict).

Check assignment

Pilot 1: $\{1, 2, 5, 6\}$ , Pilot 2: $\{3, 4\}$ . The high-contamination pairs $(1,3)$ and $(2,4)$ are on different pilots — contamination between them eliminated. The residual contamination is $\rho_{1,2}$ (intra-cell, always 0 by orthogonal intra-cell pilots in single-cell), $\rho_{1,5}, \rho_{1,6}, \rho_{2,5}, \rho_{2,6}$ — all small by assumption.

SINR vs. Pilot Reuse Factor

Compare per-user SINR for different pilot reuse strategies: universal reuse (all cells share the same pilot pool), partial reuse (pilot pools are partially disjoint), and fractional reuse. Tradeoffs appear between spectral efficiency per user and pilot overhead.

Parameters

Pilot reuse factor

f_p

1

Number of cells

L

7

BS antennas

N_t

64

Users per cell

K

10

Pilot Assignment Algorithm Comparison

Method	Complexity	Optimality	Requires $\mathbf{R}_k$ ?	Scalable?
Random	$\mathcal{O}(1)$	Poor	No	Yes
Greedy (sum contamination)	$\mathcal{O}(K^2\tau_p)$	Near-optimal in practice	Yes	Yes
Graph coloring	$\mathcal{O}(K^2 + K\|E\|)$	Good structural guarantee	Yes (threshold)	Yes
Optimal (exhaustive)	$\mathcal{O}(\tau_p^K)$	Optimal	Yes	No ( $K > 15$ )
Hungarian algorithm (matching)	$\mathcal{O}(K^3)$	Optimal for 2-cell	Yes	Moderate

,

Common Mistake: Pilot Assignment Must Be Adaptive

Mistake:

Once pilots are assigned at system startup, they can remain fixed indefinitely.

Correction:

Optimal pilot assignment depends on the covariance matrices $\mathbf{R}_k$ , which evolve as users move. A user who was at the cell edge (high contamination) may move to a region with a very different angular spectrum — changing its contamination profile with neighboring cells.

Pilot assignment should be updated whenever the large-scale fading coefficients change significantly, typically every 10–100 ms for pedestrian users and every 1–5 ms for vehicular users. In practice, 5G NR supports semi-persistent SRS configurations that can be updated by RRC reconfiguration, but the mechanism for dynamic pilot assignment based on covariance estimates is still an active research area.

🔧Engineering Note

SRS Resource Management in 5G NR

In 5G NR, the SRS (sounding reference signal) plays the role of uplink pilots for channel reciprocity estimation. The gNB assigns SRS resources (time-frequency location within the SRS occasion) to UEs via RRC configuration. Unlike the academic models studied here, 5G NR does not currently include a standardized algorithm for covariance-aware pilot (SRS) assignment — this is left to gNB implementation.

Commercial gNBs typically assign SRS resources based on cell load and UE scheduling priority rather than spatial covariance structure. Research shows 1–3 dB gain from covariance-aware assignment in dense urban scenarios, but this requires accurate covariance estimation and frequent updates.

Practical Constraints

•
5G NR SRS: up to 8 orthogonal SRS sequences per resource set (TS 38.211, Section 6.4.1.4.2)
•
Pilot reuse across cells not specified in 3GPP — implementation-specific
•
Covariance matrix dimensions scale as $N_t^2$ : at $N_t = 128$ , storing $\mathbf{R}_k$ costs 256 KB per user

📋 Ref: 3GPP TS 38.211, Section 6.4.1.4

Key Takeaway

Good pilot assignment can raise the contamination floor significantly. By assigning pilots so that users with heavily overlapping spatial covariance matrices receive orthogonal pilots, contamination is shifted to user pairs with nearly orthogonal spatial signatures. Greedy and graph-coloring algorithms provide practical, near-optimal assignment at polynomial cost. However, assignment alone cannot eliminate contamination entirely when the total user count exceeds the pilot pool — decontamination via covariance subspace structure (Section 5) is needed.

Pilot Assignment Algorithms