Wireless 101
M10 · L02
Module 10 · Lesson 2

One Fast Stream, or a Thousand Slow Ones

M8-L3 measured the damage multipath does and then promised OFDM would fix it. This is that lesson — and the fix turns multipath from a hazard into one complex multiply.

01 / 11
Wireless 101
M10 · L02
The problem, restated

Five Symbols of Interference

  • Urban delay spread στ ≈ 1 µs (M8-L3)
  • 5 MBd single carrier → Ts = 1/5×10⁶ = 200 ns (M6-L2)
  • 1 µs / 200 ns = 5 symbols of ISI → a 5-tap equaliser
  • At 50 MBd it is 50 symbols. That is the wall

Split the stream into N slow subcarriers instead. At N = 128 each symbol lasts 25.6 µs, so the echo is 3.9% of one symbol, not five whole ones.

02 / 11
Wireless 101
M10 · L02
Overlapping, yet separable

Orthogonality Costs Only Exactness

The orthogonality condition
\Delta f = \frac{1}{T_u} \;\Longrightarrow\; \int_0^{T_u} e^{\,j2\pi (k-l)\Delta f t} dt = 0

At Δf = 15 kHz and Tu = 66.67 µs, subcarrier 1200 fits exactly 1200.0 cycles. Their sinc spectra overlap heavily — and each is zero at every other centre.

03 / 11
Wireless 101
M10 · L02
Why 1966 had to wait for 1995

The IFFT Is the Modulator

  • N oscillators and N filters, or one transform
  • Direct sum, N = 2048: N² = 4,194,304 multiplies
  • Radix-2 FFT: (N/2)log₂N = 1024 × 11 = 11,264
  • Ratio 372×, for a bit-identical answer
  • Chang 1966, Weinstein 1971, CP 1980 — shipped 1995

OFDM’s adoption date was set by Moore’s law, not by insight. Nobody doubted the mathematics; nobody could afford the arithmetic.

04 / 11
Wireless 101
M10 · L02
The most elegant idea here

Paste the Tail on the Front

Overhead, and the design rule
\eta_{CP} = \frac{T_{CP}}{T_u + T_{CP}}, \quad T_{CP} > \sigma_\tau

An echo inside the prefix becomes a cyclic shift, and a cyclic shift is a phase rotation. So Yk = HkXk: one complex multiply per subcarrier replaces the equaliser. LTE’s 4.69 µs covers 1.4 km of excess path.

05 / 11
Wireless 101
M10 · L02
Try it — 20 MHz total

Symbol, Prefix, Echo

FFT N 64
CP 0.80 µs
Delay sp 100 ns
Δf 312.5 kHz Tu 3.20 µs CP overhead 20.0% Bc 2.00 MHz τmax 0.46 µs echo within CP — no ISI

The prefix has to cover the last echo, not the RMS spread στ. For an exponential delay profile truncated 20 dB down that is τmax = 4.6 στ — computed from that model, and the 20 dB is the one chosen number. Real profiles agree: ITU-R M.1225 Vehicular A has στ = 370 ns with its last tap at 2510 ns, 6.8×.

06 / 11
Wireless 101
M10 · L02
Two numerologies, verified

LTE and 802.11a

  • LTE: Δf 15 kHz → Tu = 1/15000 = 66.67 µs
  • CP 4.69 µs → 4.69/71.36 = 6.6% overhead
  • WiFi: 20×10⁶/64 = 312.5 kHz → Tu = 3.2 µs
  • CP 0.8 µs → symbol 4.0 µs, 0.8/4.0 = 20%
  • Only 48 of 64 subcarriers carry data: 48/64 = 75%

Check: 48 / 4.0 µs × 6 bits × 3/4 = 54 Mbit/s ✓ — 802.11a’s top rate, rebuilt from the numerology.

07 / 11
Wireless 101
M10 · L02
The bill, honestly

What OFDM Costs

Peak-to-average power ratio
\mathrm{PAPR} \le N \;\Rightarrow\; 10\log_{10}(2048) = 33.1\,\text{dB}

The bound is N; practice is 8 to 12 dB — still enough that LTE’s uplink uses SC-FDMA to spare the handset’s amplifier (M5-L4). Add frequency-offset sensitivity and 6.6–20% of prefix.

08 / 11
Wireless 101
M10 · L02
OFDM + multiple access

A Two-Dimensional Grid

  • Give each user a rectangle of subcarriers × symbols
  • FDMA and TDMA at once — no guard bands, no guard times
  • LTE resource block: 12 × 15 kHz = 180 kHz, 0.5 ms
  • Scheduler gives each user the subcarriers where they are strong
LTE RB
180 kHz
WiFi 6 RU
2.03 MHz
NR at 30 kHz
360 kHz
09 / 11
Wireless 101
OFDM

Check what the subcarriers do

Four questions on why OFDM splits a wide channel and what the cyclic prefix buys.

Question 1 of 0
Score 0/0

10 / 11
Wireless 101
M10 · L02
Recap

What you learned

  • N slow subcarriers turn 5 symbols of ISI into a fraction of one
  • Δf = 1/Tu makes overlapping spectra separable
  • The IFFT is the modulator — 372× cheaper at N = 2048
  • CP > στ, and 4.69/71.36 = 6.6% is what it costs LTE
  • OFDMA schedules the grid — and beats FDMA, TDMA and CDMA
Up next in Module 10
11 / 11