Wireless 101
M10 · L04
Module 10 · Lesson 4

A Call Is Two Conversations

This module shared spectrum among users. One question is left: how do the two directions of a single link share it? Only two answers are in commercial use.

01 / 11
Wireless 101
M10 · L04
Clear this up first

FDD Is Not FDMA

  • Multiple access divides a channel among users (M10-L1)
  • Duplexing divides it between the two directions of one link
  • GSM — TDMA for users, FDD for directions
  • LTE TDD — OFDMA for users, TDD for directions

Both pairings are deployed, so the two ideas are independent. Ask what is on the other side of the divide: another user, or the same user talking back?

02 / 11
Wireless 101
M10 · L04
Half, full, and the missing one

123 Decibels of Your Own Voice

Why in-band full duplex is not shipping
P_{tx} - P_{rx} = 23 - (-100) = 123\ \text{dB}

FDD and TDD both give the user full duplex — TDD just alternates faster than a conversation can notice. Same frequency, same instant needs 100+ dB of self-cancellation: a lab result, not a product.

03 / 11
Wireless 101
M10 · L04
FDD — worked on LTE Band 1

Two Bands and a Duplexer

  • Uplink 1920–1980 MHz → 1980 − 1920 = 60 MHz
  • Downlink 2110–2170 MHz → also 60 MHz
  • Duplex spacing 2110 − 1920 = 190 MHz
  • Duplex gap 2110 − 1980 = 130 MHz, and 190 − 60 = 130 ✓
  • Band 3 is tighter: 1805 − 1785 = 20 MHz of gap

The gap is the filter’s room to fall. A duplexer gives 50–60 dB of isolation, costs 1–2 dB, and there is one per band.

04 / 11
Wireless 101
M10 · L04
TDD — 300 m per microsecond

One Band and a Stopwatch

Guard period, and its overhead
T_g \ge \frac{2R}{c} \;\Rightarrow\; \eta_g = \frac{2R}{c\,T_{sw}}
  • 10 km cell: 2 × 10 000/(3×10⁸) = 66.7 µs = 1.33% of 5 ms
  • 30 km: 200 µs (4.0%). 100 km: 666.7 µs (13.3%)
  • One LTE symbol of guard, 71.4 µs, reaches 10.7 km
05 / 11
Wireless 101
M10 · L04
Try it — 5 ms switching, 2 GHz

FDD vs TDD, Side by Side

Scheme FDD
DL:UL 4 : 1
Radius 10 km
Speed 5 km/h
usable 62.5% asym waste 37.5% guard none Tc 45.7 ms no reciprocity — feedback needed
06 / 11
Wireless 101
M10 · L04
TDD’s biggest win

A Fixed Split Wastes 37.5%

What 50:50 costs against r:1 traffic
\text{waste} = 1 - \tfrac{1}{2}\!\left(1+\tfrac{1}{r}\right) = \frac{r-1}{2r}

At 4:1 demand, FDD serves 0.5 down + 0.125 up = 0.625, so (4−1)/8 = 37.5% idles ✓. TDD just asks for 4:1 and pays 1.33% of guard. At 1:1, FDD wins — which is why voice networks were FDD.

07 / 11
Wireless 101
M10 · L04
Why massive MIMO chose TDD

Reciprocity, and Its Expiry Date

The condition reciprocity requires
T_{sw} \ll T_c \approx \frac{0.423\,\lambda}{v}
Walking
46 ms
100 km/h
2.3 ms
Switching
5 ms

One frequency means the uplink measures the downlink for free — gold for M10-L3’s arrays. But M8-L3’s Tc is 2.3 ms at 100 km/h: stale before use.

08 / 11
Wireless 101
M10 · L04
Where each one lives

Below 3 GHz, FDD

  • FDD — bands 1, 3, 8; satellite, whose 239 ms round trip rules TDD out
  • TDD — n41, n77–n79 at 2.5 and 3.5 GHz; unpaired and asymmetric
  • TDD everywhere in mmWave — n257–n261, 26 and 28 GHz
  • WiFi has been TDD-like since 1997 — CSMA/CA on one channel
  • TDD needs synchronisation: an unaligned neighbour is an outage (M9-L3)

5G NR supports both. The band plan decides, and a band cannot be converted.

09 / 11
Wireless 101
Duplexing

Check what splits the directions

Four questions on FDD versus TDD and what each split costs.

Question 1 of 0
Score 0/0

10 / 11
Wireless 101
M10 · L04
Recap

What you learned

  • Duplexing splits directions; multiple access splits users
  • Band 1: 60 MHz each, 190 MHz spacing, 130 MHz duplex gap
  • Guard = 2R/c → 66.7 µs at 10 km, 1.33% of 5 ms
  • Fixed 50:50 wastes (r−1)/2r — 37.5% at 4:1 traffic
  • Reciprocity needs Tsw ≪ Tc; TDD needs synchronisation
Module 10 complete — up next
Module 11: Standards →
All lessons Read article
11 / 11