Ferkans — Interactive Telecom Tutor

Adapting to the Channel

The scheduler decides which user to serve and which resources to assign; link adaptation decides how to transmit — specifically, which modulation order and code rate to use. If the modulation-and-coding scheme (MCS) is too aggressive for the current channel, the block error rate (BLER) is unacceptably high and throughput collapses. If it is too conservative, spectral efficiency is wasted. Adaptive modulation and coding (AMC) closes the loop between channel measurement, MCS selection, and HARQ feedback, enabling the physical layer to track fading and operate near capacity in every time slot.

Definition:
Adaptive Modulation and Coding (AMC)

Adaptive modulation and coding dynamically selects the modulation order $M_{\text{mod}}$ and forward error correction (FEC) code rate $R_c$ based on the instantaneous channel quality. The combination $(M_{\text{mod}}, R_c)$ is called the modulation and coding scheme (MCS).

The spectral efficiency of MCS index $i$ is:

$\eta_i = R_{c,i} \cdot \log_2 M_{\text{mod},i} \quad [\text{bits/s/Hz}]$

AMC selects the highest MCS $i^{\star}$ such that the predicted BLER satisfies:

$\text{BLER}(i^{\star}, \gamma) \leq \text{BLER}_{\text{target}}$

where $\gamma$ is the reported post-equalisation SINR and $\text{BLER}_{\text{target}}$ is typically 10% for initial transmission (with HARQ providing residual error correction).

In LTE and 5G NR, the MCS table contains 29--32 entries ranging from QPSK rate-1/8 ( $\eta \approx 0.25$ bits/s/Hz) to 256-QAM rate-0.93 ( $\eta \approx 7.4$ bits/s/Hz). The UE reports a Channel Quality Indicator (CQI) that maps to the highest supportable MCS.

,

Definition:
Channel Quality Indicator (CQI) and MCS Table

The CQI is a quantised representation of channel quality reported by the UE to the base station. In LTE and 5G NR:

The UE measures the reference-signal received power (RSRP) and computes the post-equalisation SINR $\gamma$ per RB or subband.
The UE selects the highest CQI index $q$ such that a hypothetical transport block using the corresponding MCS would have BLER $\leq 10\%$ .
The base station uses the CQI to select the actual MCS from the MCS table:

CQI	Modulation	Code rate	Spectral efficiency
1	QPSK	0.076	0.15 bits/s/Hz
7	16-QAM	0.369	1.48 bits/s/Hz
10	64-QAM	0.332	1.99 bits/s/Hz
15	64-QAM	0.772	4.63 bits/s/Hz

(Abridged; the full 5G NR table has 15 CQI levels and up to three MCS tables supporting 256-QAM.)

CQI reporting introduces a feedback delay of 4--8 ms, during which the channel may change. Outer-loop link adaptation (OLLA) compensates by adjusting a bias term based on HARQ ACK/NACK statistics.

,

MCS Selection and BLER Curves

Explore the BLER vs. SNR curves for different MCS levels. The horizontal line marks the target BLER; the intersection with each curve defines the minimum SNR required for that MCS. Adjusting the target BLER shifts the operating point: a lower target is more conservative (lower spectral efficiency but fewer retransmissions), while a higher target is more aggressive (higher spectral efficiency but more HARQ retransmissions).

Parameters

Target BLER0.1

Example: MCS Selection for a Given Channel

A UE reports a wideband SINR of $\gamma = 15$ dB. The BLER curves (from link-level simulations) give the following BLER at 15 dB for selected MCS indices:

MCS index	Modulation	Code rate	BLER at 15 dB
10	16-QAM	0.602	0.001
15	64-QAM	0.466	0.03
20	64-QAM	0.667	0.08
22	64-QAM	0.772	0.15
25	256-QAM	0.554	0.35

The target BLER is 10%.

(a) Determine the selected MCS. (b) Compute the spectral efficiency. (c) What is the effective throughput after accounting for retransmissions?

Solution

MCS selection

(a) Select the highest MCS with BLER $\leq 10\%$ :

MCS 25: BLER = 0.35 > 0.10 (reject)
MCS 22: BLER = 0.15 > 0.10 (reject)
MCS 20: BLER = 0.08 $\leq$ 0.10 (accept)

Selected MCS: 20 (64-QAM, rate 0.667).

Spectral efficiency

(b) $\eta_{20} = 0.667 \times \log_2(64) = 0.667 \times 6 = 4.00$ bits/s/Hz.

Effective throughput

(c) With BLER = 0.08 and one HARQ retransmission (Chase combining), the effective throughput is approximately:

$\eta_{\text{eff}} = \eta_{20} \times (1 - \text{BLER}) = 4.00 \times 0.92 = 3.68 \;\text{bits/s/Hz}$

If we had chosen MCS 22 (BLER = 0.15): $\eta_{22} = 0.772 \times 6 = 4.63$ , $\eta_{\text{eff}} = 4.63 \times 0.85 = 3.94$ bits/s/Hz.

Despite the higher BLER, MCS 22 would have given better effective throughput. This illustrates why some systems use a target BLER of 10--15% and rely on HARQ to recover errors. $\blacksquare$

Quick Check

Why do LTE and 5G NR typically set the initial transmission target BLER to 10% rather than a much lower value like 0.1%?

Because a 10% BLER maximises spectral efficiency when combined with HARQ retransmissions

Because lower BLER targets are impossible to achieve

Because 10% is the minimum BLER that the decoder can achieve

Because CQI feedback is too inaccurate for lower targets

Correction:

Because a 10% BLER maximises spectral efficiency when combined with HARQ retransmissions

A higher target BLER allows a more aggressive (higher spectral efficiency) MCS to be selected. The occasional errors are corrected by HARQ with incremental redundancy, whose combining gain typically more than compensates for the initial error. The overall throughput $(1-\text{BLER}) \times \eta$ is maximised at a BLER around 10% for most channel conditions.

⚠️Engineering Note

Outer-Loop Link Adaptation (OLLA)

CQI reports are inherently imperfect due to:

Feedback delay: 4--8 ms between measurement and use, during which the channel changes.
Quantisation: CQI is a 4-bit index (15 levels), introducing up to 1--2 dB granularity error.
UE implementation bias: Different UE chipsets may map the same channel to different CQI values.

OLLA compensates by maintaining a per-UE bias $\Delta_k$ that shifts the CQI-to-MCS mapping:

$\text{MCS}_k = f(\text{CQI}_k + \Delta_k)$

The bias is updated after each HARQ outcome:

$\Delta_k \leftarrow \Delta_k + \begin{cases} \delta_{\text{up}} & \text{if ACK (step-up)} \\ -\delta_{\text{down}} & \text{if NACK (step-down)} \end{cases}$

With $\delta_{\text{down}} / \delta_{\text{up}} = \text{BLER}_{\text{target}} / (1 - \text{BLER}_{\text{target}}) \approx 1/9$ for 10% target BLER, the bias converges to a value that drives the actual BLER toward the target. OLLA is simple, robust, and universally deployed in LTE and 5G NR base stations.

Practical Constraints

•
OLLA convergence requires $\sim$ 50--100 HARQ outcomes per UE — slow for low-traffic users
•
Step sizes must be tuned: too large causes oscillation, too small tracks slowly
•
OLLA cannot compensate for sudden channel changes (e.g., handover); it must be reset

🔧Engineering Note

Scheduling Complexity at Scale

The PF scheduler must evaluate $K$ users every scheduling interval. In 5G NR with 1 ms slots and $K = 500$ connected UEs:

CQI processing: Each UE reports CQI per subband (up to 18 subbands in 100 MHz). Processing $500 \times 18 = 9000$ CQI values per slot requires efficient memory access patterns.
Metric computation: $K$ divisions $R_k/\bar{T}_k$ per slot. At 1 GHz clock, this allows only 2000 cycles per user — tight for advanced metrics with QoS weighting.
HARQ management: Up to 16 parallel HARQ processes per UE, requiring state tracking for $500 \times 16 = 8000$ processes.
Grant signalling: Each scheduled UE needs a DCI (downlink control information) message, competing for limited PDCCH capacity. With 2-symbol CORESET, typically 10--20 UEs can be scheduled per slot.

These constraints explain why practical schedulers use simplified PF variants, process UEs in priority groups, and limit the candidate set rather than evaluating all $K$ UEs on every slot.

Practical Constraints

•
PDCCH capacity limits the number of simultaneously scheduled UEs to 10--20 per slot
•
Scheduler latency budget is $< 100\;\mu$ s in many implementations
•
Memory bandwidth for CQI tables dominates compute in large cells

Historical Note: From Fixed Allocation to OFDMA

2000--2009

Early cellular systems (GSM, IS-95) used fixed resource allocation: each user was assigned a dedicated time slot or code, regardless of channel conditions. The first channel-aware scheduler appeared in Qualcomm's HDR (High Data Rate) system (2000), later standardised as 1xEV-DO. Its proportional fair scheduler, based on the Viswanath– Tse–Laroia framework, demonstrated dramatic real-world gains: 2--3× sector throughput improvement over round-robin.

The jump to OFDMA in LTE (2009) added frequency-domain scheduling: not only when to serve each user, but which subcarriers to assign. 3GPP's LTE scheduler interface was designed with explicit support for per-RB CQI reporting and per-RB scheduling grants, enabling fine-grained exploitation of frequency-selective fading. This two-dimensional (time × frequency) scheduling framework remains the foundation of 5G NR resource management.

Why This Matters: CQI Tables in 5G NR

5G NR (3GPP TS 38.214) defines three CQI tables, each mapping 15 CQI indices to modulation and code-rate pairs:

Table 1 (default): QPSK to 64-QAM, for general eMBB traffic with 10% BLER target.
Table 2 (256-QAM): Extended to 256-QAM for high-SNR scenarios, yielding up to 7.4 bits/s/Hz spectral efficiency.
Table 3 (URLLC): More conservative MCS entries with $10^{-5}$ target BLER for ultra-reliable low-latency communications, sacrificing peak throughput for reliability.

The gNB signals which CQI table to use via RRC configuration. For URLLC, the combination of Table 3 CQI, mini-slot scheduling, and redundancy version 0 retransmission enables the stringent $10^{-5}$ reliability and $\leq 1$ ms latency requirements.

Adaptive Modulation and Coding (AMC)

A link adaptation technique that dynamically selects the modulation order and FEC code rate based on the instantaneous channel quality, choosing the highest-spectral-efficiency MCS that meets a target BLER. AMC is a core cross-layer mechanism in all modern cellular systems.

Channel Quality Indicator (CQI)

A quantised index reported by the UE to the base station, indicating the highest MCS that can be supported at a target BLER (typically 10%). CQI drives the link adaptation and scheduling decisions at the base station.

Modulation and Coding Scheme (MCS)

A combination of modulation order (e.g., QPSK, 16-QAM, 64-QAM, 256-QAM) and FEC code rate that determines the spectral efficiency of a transmission. The MCS table maps indices to specific (modulation, code rate) pairs.

Link Adaptation and AMC