Ferkans — Interactive Telecom Tutor

What Chip Does the OTFS Cost?

Standards committees care about more than algorithmic elegance — they care about whether the technology can run on reasonable silicon at reasonable cost. 3GPP's adoption of OTFS in 6G hinges on whether chip vendors can deliver OTFS-capable modems at a price competitive with enhanced-OFDM alternatives. This section quantifies the compute, memory, and power costs of OTFS relative to the 5G NR baseline.

Definition:
Baseband Compute Budget

5G NR baseband compute (per-frame, one UE-BS link, 100 MHz):

FFT: $N \log_2 N \sim 1.4 \times 10^4$ ops.
Channel estimation (DMRS): $\sim 10^5$ ops.
MIMO detection (MMSE): $\sim 10^6$ ops.
LDPC decoding: $\sim 10^6$ ops.

Total: $\sim 3 \times 10^6$ ops per frame. At 100 Hz frame rate: $\sim 3 \times 10^8$ ops/sec per link.

OTFS baseband compute:

ISFFT/SFFT: $MN \log_2(MN) \sim 10^5$ ops.
DD-domain channel estimation: $\mathcal{O}(P \cdot MN)$ $\sim 10^5$ ops.
MP detection: $\sim 3 \times 10^6$ ops (per Chapter 8).
Decoding: same as 5G NR.

Total: $\sim 5 \times 10^6$ ops per frame. ~ $1.5\times$ over 5G NR.

Theorem: OTFS Compute Scaling

For 5G NR + OTFS dual-mode UE at 100 MHz bandwidth, per-frame compute is $C_{\mathrm{OTFS}} \;\leq\; 1.5 \cdot C_{\mathrm{5G NR}}.$ The factor $1.5\times$ comes from:

OTFS ISFFT: adds $\mathcal{O}(MN \log MN)$ per frame.
MP detection: vs linear MMSE, factor $\sim 1.5\times$ at convergence.
DD channel estimation: $\sim 1.2\times$ vs standard DMRS.

Consequence: a dual-mode UE with OTFS support costs $\sim 50\%$ more baseband compute than 5G NR-only. In silicon terms: $\sim 20\%$ more area, $\sim 30\%$ more power. Acceptable for 2028+ semiconductor nodes.

The extra compute is the price of DD-domain processing. Not huge in percentage terms, but significant at the chip level — especially for battery-powered UEs. The saving grace: modern semiconductor nodes (TSMC 3nm, 2nm) easily absorb this overhead at no meaningful price increase. Mass-market 6G chips can support OTFS natively.

Proof

Per-operation cost

ISFFT/SFFT: 2D FFTs, $\mathcal{O}(MN \log MN)$ complex multiplies. MP detection: iterative, $\sim 5$ - $10$ iterations × $\mathcal{O}(MN P)$ per iter.

Compare to NR

5G NR: FFT per OFDM symbol + per-subcarrier MMSE. For same aggregate bandwidth: $\mathcal{O}(N \log N)$ per symbol × slots/frame.

Ratio

OTFS / NR: $\sim 1.5\times$ for realistic $P$ .

Memory

DD channel matrix: $\sim MN \cdot N_r N_t$ elements. For typical: $\sim 10^4$ complex. Modest.

Silicon implications

TSMC 3nm: 20% area increase at 50% compute increase. Power: 30% at similar gate. Mass-market UE cost: +~ $20-30.$ \blacksquare$

Key Takeaway

OTFS's 1.5× compute cost is absorbable. Modern semiconductor nodes (3nm, 2nm, 1.4nm) deliver enough transistor density that the extra compute is $\sim 20\%$ area, $\sim 30\%$ power — minor in mass-market 6G chips. Compute is not the barrier.

Definition:
OTFS Memory Requirements

OTFS memory footprint (per active session):

DD channel matrix $\mathbf{H}_{DD} \in \mathbb{C}^{MN \times MN}$ (full): $\mathcal{O}(MN)^2$ complex. For $MN = 10^4$ : $10^8$ complex $= 8$ GB. Impractical.
Sparse DD channel: only $P$ nonzero entries. $8 P$ bytes. For $P = 10$ : 80 bytes. Trivial.
Received frame: $MN$ complex samples. $10^5$ bytes (100 KB).
Transmit frame: same.
Detection state (MP messages): $MN \cdot$ iterations $\cdot P$ . $\sim 10^6$ bytes (1 MB).

Per-session total: $\sim 1-2$ MB. Per UE with multiple active sessions: $\sim 10$ - $30$ MB.

BS-side (serving many UEs simultaneously):

Per-UE: 1-2 MB.
1000 active UEs: 1-2 GB.

Modern 5G gNB RAM: 100+ GB. Easily absorbed.

Example: OTFS Chip Budget: UE vs BS

Estimate silicon cost and power of dual-mode (OFDM+OTFS) UE and BS chips for 2028-era 6G deployment.

Solution

UE

Baseband: ~2 GFLOPS at 2 GHz. Silicon area: ~20 mm² at 3nm. Cost per chip: ~$5 (mass production). Power: ~0.5 W baseband. Compared to 5G-only UE: +15% cost, +10% power.

BS

Multi-UE baseband: ~2 TFLOPS at 5 GHz. Silicon area: ~200 mm² (multi-chip). Cost: ~$2000 (low-volume production). Power: ~100 W. Compared to 5G-only BS: +10% cost, +10% power.

Deployment impact

Incremental cost of 6G OTFS over 5G baseline: ~ $20 per UE, ~$ 500 per BS. Negligible in total system cost. Expect dual-mode 6G deployments by 2030.

OTFS vs 5G NR Compute Breakdown

Stacked bar chart showing compute breakdown per frame: FFT, channel estimation, detection, decoding. Compares 5G NR, OTFS, dual-mode.

Parameters

\log_2 MN

12

P

8

Theorem: OTFS Power Consumption Trade-Off

For a dual-mode UE comparing OTFS and OFDM: $P_{\mathrm{OTFS}} \;=\; P_{\mathrm{OFDM}} + \Delta P_{\mathrm{baseband}} + \Delta P_{\mathrm{PA}},$ where $\Delta P_{\mathrm{baseband}} \sim 0.1$ W (OTFS extra compute) and $\Delta P_{\mathrm{PA}} \sim 0.5$ W (2 dB PAPR penalty → higher PA back-off, less efficient).

Consequence: OTFS UE consumes $\sim 30\%$ more power than OFDM UE at same link conditions. For battery-powered devices (IoT, handheld), this is a real cost.

Trade-off: in high-mobility scenarios, OTFS's link reliability saves power by eliminating retransmissions (which cost more than the PAPR penalty). Net: OTFS may be more efficient under mobility. At static conditions: OFDM is more efficient.

The energy cost of OTFS has two components: compute and PA. The compute cost is small. The PA cost is real — 2 dB higher PAPR means less effective Tx power. But: under mobility, OFDM's retransmissions eat power at a rate that exceeds the OTFS PAPR penalty. Net result: OTFS is energy-optimal for mobility, OFDM for static. Another argument for dual-waveform.

Proof

Compute power

OTFS: 30% more operations at same voltage/freq → 30% more power. Static power: ~0.5 W → 0.65 W.

PA power

OTFS PAPR: 7 dB. OFDM: 5 dB (DFT-s). Difference: 2 dB. PA efficiency at 2 dB back-off: 30% vs 45% (linear). Output power 20 dBm. Input PA power: $100$ mW $/0.3 = 333$ mW vs $100/0.45 = 222$ mW. Difference: 111 mW.

Retransmission

OFDM at high mobility: $\sim 20\%$ retransmission rate. Each retransmission: 500 mW × time. Aggregate: +100 mW average. OTFS: $< 1\%$ retransmission. +1 mW average.

Net

Static: OTFS +200 mW (PAPR dominates). Mobile: OTFS -150 mW (retransmission savings). Waveform selection depends on mobility. $\blacksquare$

Definition:
Silicon Roadmap for OTFS

OTFS silicon development timeline:

2024-2026: Research prototypes. FPGA-based OTFS demonstrators. Cohere Technologies + select universities. Baseband on general- purpose DSP.

2026-2028: ASIC prototypes. Integrated ISFFT + MP detection in custom silicon. 7nm node. ~ $500 per chip in small quantities. **2028-2030**: Mass-production ASICs. Dual-mode 5G NR + OTFS. 3nm node. ~$ 10 per chip at volume. First 6G UE launches.

2030+: Second-generation dual-mode. 2nm node. ~$$5 per chip. Mass consumer deployment.

Vendor landscape: Qualcomm, MediaTek, Samsung LSI expected to lead OTFS silicon. Intel (handset discontinued), Huawei HiSilicon (sanctioned), ARM (IP provider) in supporting roles.

OTFS vs 5G NR Complexity Summary

Metric	5G NR (OFDM)	6G OTFS (dual)	Ratio
Baseband compute	3 GFLOPS	5 GFLOPS	1.67×
Memory (per session)	0.5 MB	1.5 MB	3×
UE silicon area	50 mm²	60 mm²	1.2×
UE chip cost (volume)	$5	$8	1.6×
UE battery drain	Baseline	+10-15%	1.1×
BS silicon area	150 mm²	180 mm²	1.2×
BS power	150 W	200 W	1.33×
Standardization readiness	2020	2028-2030	—

🔧Engineering Note

Vendor Ecosystem and OTFS Adoption

OTFS commercial adoption depends on vendor ecosystem readiness:

Chip vendors:

Qualcomm: active OTFS research (~2020+). Expect OTFS-capable 6G chips by 2028.
MediaTek: OTFS research in 2024+. Likely 2029+ mass production.
Samsung LSI: same timeline.
Cohere Technologies: OTFS IP owner. Licensing to semiconductor vendors.

Infrastructure:

Ericsson, Nokia, Samsung: OTFS study participation in 3GPP. Commercial BS 2029+.
Huawei (China market): OTFS development ongoing. Commercial 2028+.
Mavenir, Rakuten (open RAN): may lead OTFS adoption for greenfield deployments.

Operators:

T-Mobile US, Verizon: likely 6G OTFS deployers. 2030+.
Japan, Korea, China: lead markets.
Europe: conservative; 6G OTFS 2031+.

Commercial readiness: aligned with Rel. 21 (2028-2030) specification and Rel. 22 (2032+) mass deployment.

Practical Constraints

•
Chip vendors: Qualcomm, MediaTek lead OTFS silicon
•
Cohere Technologies: OTFS IP provider
•
Infrastructure: Ericsson, Nokia, Samsung active in 3GPP OTFS
•
Commercial: 2028+ pilots, 2030+ mass

Common Mistake: The Chicken-and-Egg Problem

Mistake:

Assuming vendors will build OTFS chips once standards finalize. In reality, vendors need business cases — operators need equipment — which needs vendor investment — which needs operator commitment. A chicken-and-egg deadlock.

Correction:

The CommIT contributions of this book (Chapters 17-18) are the kind of quantitative case that breaks the deadlock: concrete scenarios (cell-free, LEO) where OTFS delivers measurable gains ( $35\%$ throughput, 100 $\times$ reliability). Operators see these and commit; vendors respond. The standardization follows. Expect first committed deployments 2028-2029.

Implementation Complexity Assessment