The transition from NRZ to PAM4 signaling has changed the way high-speed optical transceivers process electrical and optical signals. As Ethernet interfaces move toward 100G, 400G, 800G, and 1.6T, increasing bandwidth per lane requires more sophisticated modulation and receiver architectures.
Traditional NRZ links often use CDR-based receiver designs to recover clock timing and retime incoming data. Higher-speed PAM4 links generally require more advanced signal processing, including multi-level detection, equalization, and sophisticated clock recovery. DSP-based architectures are widely used to manage these requirements.
The difference is not simply that PAM4 uses DSP while NRZ uses CDR. CDR is a timing-recovery function that can be integrated into either architecture. The main engineering distinction is the additional signal-processing complexity associated with four-level signaling.
1. What Is PAM4 Signaling?
PAM4 stands for four-level Pulse Amplitude Modulation. It uses four distinct signal amplitude levels to represent digital information. Each symbol carries two bits, allowing PAM4 to transmit twice as many bits per symbol as NRZ.
The four signal levels are commonly represented as 00, 01, 10, and 11, although the actual Gray-code mapping and level assignment depend on the implementation.
For example, a PAM4 signal can represent two bits during one symbol interval instead of transmitting one bit per interval as in NRZ. This increases data capacity without requiring the same increase in symbol rate.
Four amplitude levels.
Two bits per symbol.
Higher bandwidth efficiency than NRZ at the same bit rate.
Smaller voltage or optical-power separation between adjacent levels.
Greater sensitivity to noise, distortion, and signal-level variations.
The smaller spacing between PAM4 levels makes accurate signal detection more challenging. Receiver designs therefore require careful control of equalization, sampling, timing recovery, and error performance.
2. What Is NRZ Signaling?
NRZ stands for Non-Return-to-Zero. It is a binary modulation format that uses two signal levels to represent digital 0 and 1. Each symbol carries one bit of information.
Because NRZ has only two amplitude levels, the receiver distinguishes between a high and a low signal level. This relatively simple detection process has supported many generations of Ethernet, storage, and optical communication interfaces.
Two signal amplitude levels.
One bit per symbol.
Larger signal-level separation than PAM4 under comparable normalized conditions.
Relatively straightforward clock and data recovery.
Higher symbol rate than PAM4 for the same bit rate.
At higher data rates, NRZ requires increasingly wide electrical and optical bandwidth. Channel attenuation and inter-symbol interference can become major limitations, particularly when electrical traces, connectors, and optical components operate close to their bandwidth limits.
3. PAM4 DSP vs NRZ CDR: Main Technical Differences
PAM4 DSP and NRZ CDR describe different aspects of a transceiver architecture. PAM4 identifies the modulation format, while DSP describes a broad set of digital signal-processing functions. CDR is a specific function that recovers timing and retimes data.
| Comparison | PAM4 DSP-Based Architecture | NRZ CDR-Based Architecture |
|---|---|---|
| Signal format | Four-level PAM4 | Two-level NRZ |
| Bits per symbol | 2 bits | 1 bit |
| Signal detection | Requires distinguishing four amplitude levels | Distinguishes two amplitude levels |
| Equalization | Often uses advanced DSP-based equalization | May use analog equalization and simpler receiver processing |
| Clock recovery | Requires PAM4-aware timing recovery | Uses binary signal timing recovery |
| Signal margin | Smaller separation between adjacent levels | Greater separation between the two levels under comparable normalized conditions |
| Receiver complexity | Generally higher | Generally lower for comparable receiver implementations |
| Power consumption | Can be higher due to advanced processing requirements | Can be lower with simpler receiver architectures |
| Typical applications | High-speed 50G-per-lane and above interfaces | Traditional 10G, 25G and selected 100G lane architectures |
| FEC considerations | Commonly requires FEC in high-speed Ethernet implementations | Depends on the interface standard and system architecture |
The table describes common implementation trends, not fixed rules. NRZ transceivers can use DSP, and PAM4 receivers can use analog or hybrid processing. The actual architecture depends on the data rate, reach, electrical channel, and applicable interface standard.
4. Why PAM4 Requires More Advanced Signal Processing
The main challenge with PAM4 is its reduced signal-level spacing. Four amplitude levels must be accurately distinguished even when the received waveform contains noise, distortion, jitter, and inter-symbol interference.
Compared with binary NRZ signaling, PAM4 has more possible transitions between adjacent and non-adjacent levels. The receiver must therefore evaluate both signal amplitude and transition timing with greater precision.
A PAM4 DSP may perform several important functions:
Feed-Forward Equalization (FFE): Compensates for linear channel distortion and reduces inter-symbol interference.
Decision Feedback Equalization (DFE): Uses previous symbol decisions to compensate for selected post-cursor interference.
Clock and Data Recovery: Extracts timing information and determines suitable sampling positions.
Signal-level detection: Processes the received waveform to distinguish among the four PAM4 levels.
Digital filtering: Helps manage signal distortion and improve receiver performance.
These functions can be implemented in different combinations of digital and analog circuitry. A DSP-based receiver may use an ADC and digital processing, while other designs rely on analog front-end equalization and specialized timing-recovery circuits.
5. How NRZ CDR Recovers Clock and Data
Clock and Data Recovery is responsible for extracting timing information from an incoming serial data stream. The receiver uses signal transitions to estimate the clock frequency and phase, allowing it to sample the data at appropriate points.
In a traditional NRZ receiver, CDR typically performs the following operations:
Detects signal transitions in the incoming binary waveform.
Estimates the timing phase and frequency.
Adjusts the recovered sampling clock.
Samples the incoming data near the center of the data eye.
Retimes the recovered data for the next processing stage.
CDR helps reduce the impact of timing variations and jitter within its operating range. However, it does not provide the full range of equalization and signal-reconstruction capabilities associated with an advanced DSP.
Incoming NRZ Signal → Analog Front-End / Equalization → CDR → Data Retiming → Host Interface
This is a simplified architecture. Actual implementations may include CTLE, DFE, limiting amplifiers, and additional signal-conditioning functions.
6. PAM4 DSP and NRZ CDR: Equalization Requirements
Equalization is essential in high-speed serial communication because electrical and optical channels can distort the transmitted waveform. Frequency-dependent attenuation, reflections, crosstalk, and bandwidth limitations can cause adjacent symbols to interfere with one another.
The required equalization complexity depends on the modulation format, data rate, channel loss, and receiver architecture.
| Equalization Function | PAM4 DSP Architecture | NRZ CDR Architecture |
|---|---|---|
| CTLE | May be used in the analog front end | Commonly used to compensate for high-frequency loss |
| FFE | Often implemented as part of DSP processing | May be implemented in analog or digital circuitry |
| DFE | Can improve detection under selected channel conditions | Used in some high-speed NRZ receivers |
| Clock recovery | Requires timing recovery compatible with PAM4 transitions | Recovers timing from binary transitions |
| Signal detection | Four-level decision process | Two-level decision process |
PAM4 equalization is particularly important because smaller vertical eye openings leave less margin for noise and distortion. NRZ also requires equalization at high speeds, but its binary detection structure generally provides a simpler signal-decision problem.
7. Clock Recovery Challenges in PAM4 and NRZ
CDR is required in both PAM4 and NRZ receivers, but the timing-recovery problem is more complex with PAM4 signaling.
In NRZ, transitions occur between two amplitude levels. The receiver can use these transitions to estimate timing and adjust the sampling clock.
PAM4 has four amplitude levels and multiple possible transitions. Some transitions occur between adjacent levels, while others span multiple levels. These transitions can have different amplitudes and slopes, making timing extraction more challenging.
| CDR Consideration | NRZ | PAM4 |
|---|---|---|
| Amplitude levels | 2 | 4 |
| Transition types | Two-level transitions | Multiple transition amplitudes |
| Timing information | Derived from binary transitions | Derived from multi-level transition behavior |
| Phase detection | Relatively straightforward binary timing detection | Requires PAM4-aware phase detection |
| Receiver design | Established analog CDR architectures are widely used | May use specialized analog, digital, or hybrid CDR |
PAM4 clock recovery may use baud-rate CDR, Mueller-Muller phase detection, or other timing-recovery approaches. The selected method depends on the receiver design and signal conditions.
8. Data Rate and Bandwidth: PAM4 vs NRZ
PAM4 increases the number of bits transmitted per symbol, which reduces the symbol rate required for a given bit rate. This is one of the main reasons PAM4 is widely used in modern high-speed Ethernet interfaces.
For the same gross bit rate, PAM4 operates at approximately half the symbol rate of NRZ. Its Nyquist frequency is therefore also approximately half, assuming comparable line-rate definitions.
| Bit Rate | NRZ Symbol Rate | PAM4 Symbol Rate | NRZ Nyquist Frequency | PAM4 Nyquist Frequency |
|---|---|---|---|---|
| 50 Gb/s | 50 GBd | 25 GBd | 25 GHz | 12.5 GHz |
| 100 Gb/s | 100 GBd | 50 GBd | 50 GHz | 25 GHz |
| 200 Gb/s | 200 GBd | 100 GBd | 100 GHz | 50 GHz |
The table shows idealized gross symbol-rate relationships before accounting for coding, FEC overhead, framing, and implementation-specific line-rate definitions. It is not a specification for a particular Ethernet interface.
The lower symbol rate of PAM4 can reduce the required channel bandwidth for a given bit rate. However, the smaller spacing between its four signal levels creates stricter signal-to-noise and linearity requirements.
9. Power Consumption and Latency
PAM4 DSP architectures generally require more complex processing than simpler NRZ CDR-based designs. This can increase transceiver power consumption and processing delay, although the actual result depends on the chip generation and implementation.
Several factors influence power consumption:
DSP processing capability and operating data rate.
ADC and DAC architecture, where implemented.
Equalization complexity and number of processing lanes.
Clock recovery and retiming circuitry.
Thermal design and power-management requirements.
Latency also varies with architecture. A DSP-based module may introduce additional processing delay through equalization, retiming, and other signal-processing stages. A simpler CDR-based design can reduce some processing stages, but it still requires clock recovery and data retiming.
| Factor | PAM4 DSP | NRZ CDR-Based |
|---|---|---|
| Processing complexity | Generally higher | Generally lower |
| Module power | Can be higher | Can be lower |
| Equalization capability | Often more advanced | Depends on the analog or digital receiver design |
| Processing latency | Depends on DSP architecture and processing pipeline | Depends on CDR and retiming implementation |
| Thermal requirements | May require more thermal management | Can have a simpler module thermal profile |
A direct comparison should use actual module specifications and measured system performance. The modulation format alone does not determine power consumption or latency.
10. FEC Requirements in PAM4 and NRZ Systems
PAM4 has smaller amplitude separation between adjacent signal levels than NRZ under comparable normalized conditions. This makes it more sensitive to noise and signal distortion.
Forward Error Correction is therefore an important part of many high-speed PAM4 Ethernet systems. FEC adds redundant information to the data stream, allowing the receiver to correct a defined number of errors.
The relationship between PAM4 DSP, CDR, and FEC can be summarized as follows:
| Technology | Primary Function | Role in the Link |
|---|---|---|
| PAM4 | Four-level modulation | Transmits two bits per symbol |
| DSP | Digital signal processing | Performs equalization, signal recovery, and other processing |
| CDR | Clock and data recovery | Recovers timing and retimes received data |
| FEC | Forward error correction | Corrects eligible transmission errors |
These functions are complementary rather than interchangeable. A PAM4 DSP can include CDR functionality, while FEC may be implemented in the host ASIC, optical module, or another part of the system.
For additional technical details, see FEC vs DSP and KP-FEC vs RS-FEC.
11. Applications in 100G, 400G, 800G, and 1.6T Networks
The choice of modulation and receiver architecture depends on the required lane rate, optical reach, host interface, and network application. NRZ remains relevant in established lower-rate interfaces, while PAM4 is widely used to increase bandwidth per lane in high-speed optical connectivity.
| Network Generation | Common Signaling Approach | Receiver Architecture Considerations | Typical Applications |
|---|---|---|---|
| 10G | NRZ | CDR and conventional equalization | Enterprise Ethernet and data center links |
| 25G | NRZ | High-speed CDR and electrical channel equalization | Server connections and 25G Ethernet |
| 100G | NRZ or PAM4 depending on lane architecture | CDR-based or DSP-based processing | Data center and optical transport |
| 400G | Commonly PAM4 | Advanced equalization, timing recovery, and FEC | Cloud data centers and AI infrastructure |
| 800G | Commonly high-speed PAM4 lanes | Advanced signal processing and channel optimization | AI clusters and high-density switching |
| 1.6T | Next-generation high-speed PAM4 architectures | Higher lane rates, improved signal processing, and power optimization | Next-generation AI and hyperscale data centers |
These are general technology trends, not mandatory configurations for every product at a given data rate. Actual modulation and receiver architecture must be verified against the applicable Ethernet standard and transceiver datasheet.
12. Selecting the Right Optical Transceiver Architecture
When selecting an optical transceiver, engineers should evaluate the complete electrical and optical link rather than choosing based only on PAM4 or NRZ signaling.
Data rate: Determine the required aggregate bandwidth and per-lane speed.
Transmission distance: Confirm the optical reach and supported fiber type.
Host interface: Verify SerDes lane rate, modulation, and electrical compatibility.
Signal processing: Identify whether the module uses DSP, CDR, or a hybrid architecture.
FEC configuration: Confirm the required FEC mode and its implementation location.
Power consumption: Evaluate module power and overall system thermal limits.
Interoperability: Validate compatibility with the target switch, router, or NIC.
For short-reach links with controlled electrical channels, a simpler receiver architecture may be suitable. For higher-speed interfaces with demanding channel conditions, advanced DSP-based processing can provide the additional equalization and signal-recovery capability required by the design.
13. C-LIGHT High-Speed Optical Transceiver Solutions
C-LIGHT provides optical transceiver solutions for data centers, enterprise networks, cloud infrastructure, and high-performance computing applications. Its portfolio includes multiple data rates and form factors designed for different optical transmission requirements.
400G QSFP-DD modules support high-bandwidth data center connectivity. Depending on the optical configuration, applications may include short-reach parallel-fiber links and longer-reach single-mode fiber connections.
Explore the C-LIGHT Optical Transceiver Portfolio for available 400G solutions.
800G OSFP solutions are designed for high-density switching, AI clusters, and next-generation data center networks. Different optical configurations support different transmission distances and network topologies.
Product page: 800G OSFP Optical Transceiver
As network bandwidth requirements increase, 1.6T optical connectivity is becoming an important direction for next-generation AI and hyperscale data center infrastructure. Module selection requires careful evaluation of lane rate, signal processing, host interface, thermal performance, and optical reach.
For available high-speed optical solutions, visit the C-LIGHT optical transceiver category.
The modulation format and internal DSP/CDR implementation can vary by product model. Customers should confirm the exact electrical interface, modulation format, FEC requirements, and compatibility specifications before deployment.
14. Conclusion
PAM4 DSP and NRZ CDR represent different approaches to high-speed signal transmission and recovery. PAM4 increases bandwidth efficiency by carrying two bits per symbol, but its four-level signal structure requires more precise amplitude detection, equalization, and clock recovery.
NRZ uses two signal levels and offers a simpler detection structure. Traditional NRZ CDR architectures remain effective in many established optical and electrical interfaces, although high-speed NRZ links can also require advanced equalization and digital processing.
For modern 400G, 800G, and 1.6T optical networks, PAM4 DSP-based processing is an important technology for managing higher lane rates and demanding signal conditions. The best architecture ultimately depends on data rate, channel quality, optical reach, power budget, latency, and host compatibility.
C-LIGHT provides optical connectivity solutions for data center, enterprise, and high-performance computing networks. Contact our team for product specifications and compatibility support.
Contact C-LIGHT
TEL:+86 132 6656 7067




















































>
>
>
>
>
>
>
>