Practical Implications for 6G D2D

From Scaling Theory to System Design

The Ji-Caire-Molisch scaling and constant analysis inform 6G architectural decisions. This section distills the practical implications for deployable D2D caching systems.

Definition:

Aggregate vs Per-User Scaling

It is important to distinguish two scaling metrics:

  1. Per-user scaling. How rate per user grows/shrinks with nn. D2D + caching: Θ(M/N)\Theta(M/N) (constant). Infrastructure: O(log⁡n/n)O(\log n/n) (shrinking).
  2. Aggregate scaling. Total network throughput: per-user⋅n\text{per-user} \cdot n. D2D + caching: Θ(nM/N)\Theta(n M/N) (linear). Infrastructure: O(log⁡n)O(\log n) (slow).

D2D aggregate throughput is superior at scale; each added user adds M/NM/N to total throughput.

Aggregate Throughput: D2D vs Cellular

D2D aggregate throughput grows linearly in nn — every added user brings cache + demand + spatial reuse opportunity. Cellular (even with MAN caching) saturates at log⁡n\log n — the shared BS capacity is the bottleneck.

Parameters
0.25
1000

Where Coded D2D Shines

Coded D2D's 3-10× constant-factor improvement is most impactful in specific regimes:

  1. Dense urban / stadium. Large nn, dense neighbors, moderate M/NM/N. Per-user throughput matters; infrastructure can't serve peak demand. Coded D2D absorbs the load.
  2. Venue-scale content delivery. Sports events, conferences, transit hubs. Users cluster geographically; popular content (α>0.5\alpha > 0.5) cached widely. Coded D2D ideal.
  3. Platooning / V2V. Vehicles within cluster share cached maps, media, telemetry. Coded gain amplifies per-vehicle rate.
  4. Wearable networks. Body-area + personal devices; small distance scales, high content correlation.

Less ideal regimes:

  1. Sparse networks (< 50 users).
  2. Unique demand (low popularity concentration).
  3. Static content (cache refresh cost low — no need for frequent coded delivery).

Example: Designing Coded D2D for a Stadium

A stadium has 50,000 spectators watching a match. Content library: live match feed (1 stream) + 100 replay / highlight files. Peak demand: each user requests some replay/highlight. Cache per phone: 10 GB. Library size: ~50 GB. Design cluster size and evaluate throughput.

Solution
Parameters

n=50,000n = 50{,}000. μ=10/50=0.2\mu = 10/50 = 0.2.

Cluster formation

Practical cluster size Kg=20K_g = 20 (few rows of seats within D2D range). Kgμ=4K_g \mu = 4.

Coded gain constant

(1+4)/(1+0.2)≈4.2(1 + 4)/(1 + 0.2) \approx 4.2. Coded D2D delivers ~4× higher per-user throughput than uncoded D2D.

Aggregate

Per-user: Θ(M/N)=0.2\Theta(M/N) = 0.2, say 20 Mbps at clink=100c_{\text{link}} = 100 Mbps. Aggregate: 50,000⋅20=10650{,}000 \cdot 20 = 10^6 Mbps = 1 Tbps. Cellular alternative: a single macro cell at 10 Gbps / 50K users = 200 kbps each. D2D gives 100× more per-user rate.

Verdict

Coded D2D is compelling for this deployment. Implementation requires 5G NR Sidelink + cluster coordination protocol.

⚠️Engineering Note

D2D Coded Delivery in 5G / 6G Standards

Implementing coded D2D delivery in standards:

  1. 5G NR Sidelink (Rel-16). Basic D2D exists; group communication (V2X) supported. Coded-caching extensions not yet standardized.
  2. ProSe (LTE/NR). Device-to-device proximity services. Supports broadcast (useful for multicast XOR) and unicast.
  3. Cache-aware protocols. Standards bodies (3GPP, ETSI) have proposals for content-addressable delivery layer. Would naturally support MAN-XOR format.
  4. 6G (Rel-19+). Cache-aided multicasting as a study item. Coded-D2D is a plausible direction.

The main gap: coordinated placement. For the coded scheme to work, users' caches need to follow a pre-designed MAN pattern. Today's CDNs use independent placement. Coordinating caches across millions of devices is an unsolved operational problem.

Practical Constraints
  • •

    5G NR Rel-16 Sidelink: unicast + group

  • •

    Cache-aware protocols: 3GPP study items only

  • •

    6G Rel-19+ study items: cache-aided multicast

  • •

    Coordinated placement: not yet standardized

Quick Check

Adding coded multicasting (MAN-style XORs) to a D2D caching network improves:

The scaling order of per-user throughput

The constant factor in per-user throughput

Both the scaling order and the constant

Nothing — coded D2D is equivalent to uncoded

Correction:
The constant factor in per-user throughput

Correct (Ji-Caire-Molisch 2015). Coded multicasting improves the constant factor (typically 2-10×) but not the scaling order. The two gain mechanisms (spatial reuse + coded multicast) do not cumulate.

The Constant Factor ImprovementChapter Summary