Five Generations, Five Different Questions
Module 10 gave you five ways to share spectrum and no history. They arrived in sequence, each one because the last had hit a wall someone could name. This is where those figures become history.
What It Solved, What It Cost
- 1G gave mobility — and no encryption, no authentication, 833 channels
- 2G gave capacity by digitising speech — and stayed a circuit
- 2.5G gave data over a voice network — and 600–700 ms of latency
- 3G made data first class — and needed power control 1500× a second
- 4G went all-IP — and lost circuit-switched voice entirely
Read a generation as a trade the industry accepted, never as a free improvement.
Eight Users, One Carrier
- Frame = 8 × 577 µs = 4.615 ms, so 217 frames/s
- 22.8 kbit/s channel carries a 13 kbit/s codec → rate 0.57
- 200/8 = 25 kHz per call vs AMPS’s 30 — only 1.2×
GSM’s real capacity win was tighter reuse, not the slots.
GPRS, Then EDGE
- Same 577 µs slots, no longer reserved — handed out on demand
- GPRS CS-4 = 21.4 kbit/s/slot → 8 × 21.4 = 171.2 kbit/s
- Real 4-slot handset on CS-2: 4 × 13.4 = 53.6 kbit/s
- EDGE adds 8PSK (M5-L3): MCS-9 = 2 × 592 bits/20 ms = 59.2 kbit/s
- 4 × 59.2 = 236.8 kbit/s advertised; 8 × 59.2 = 473.6 is the ceiling
Four coding schemes chosen per connection — M6-L4’s AMC in its crudest form.
Rate and Coverage Are the Same Resource
Run it on 384 kbit/s instead: 3.84×10⁶/384×10³ = 10.0 dB. Speech gets 25 dB of protection, data gets 10 — so data must sit closer in. Soft capacity is the reward; power control 1500×/s is the price.
Rate Grew Faster Than Efficiency
Two Routes to 7.5 bit/s/Hz
Overhead takes 201.6 down to Cat-4’s 150.75 Mbit/s, which is 74.8% of raw. Then 150/20 = 7.5 bit/s/Hz — and M6-L3 predicted exactly that from Shannon and a margin, knowing nothing of resource blocks.
They Deleted the RNC
- UMTS: NodeB → RNC → SGSN → GGSN — four hops, four queues
- LTE: eNodeB → S-GW → P-GW — and the eNodeB holds the scheduler
- MME is control plane only; S-GW/P-GW are user plane only
- ~150 ms became 30–50 ms, a factor of about 3.75
- 1 ms TTI: modulation and coding re-picked 1000×/s from CQI 1–15
Removing a hierarchy layer is what cut the latency. M6-L4’s ladder spans 36× in throughput.
LTE Cannot Make a Phone Call
No circuit domain exists, so voice is RTP/UDP/IP on QCI-1 — or CSFB drops you to 2G. 13.84 kbit/s after ROHC: almost exactly GSM’s 13.
What you learned
- 1G FDMA → 2G TDMA → 2.5G packets → 3G codes → 4G OFDMA
- 13 kbit/s to 150 Mbit/s = 11 538× in 21 years, 56%/year
- Rate ×7813 = bandwidth ×666.7 × bit/s/Hz ×11.7
- Flat EPC: no RNC, control and user planes split, 1 ms TTI
- No circuit voice at all — hence VoLTE on QCI-1, and CSFB