Muriel Médard, NEC Professor of Software Science and Engineering EECS, MIT, USA

Muriel Médard
NEC Professor of Software Science and Engineering EECS
MIT, USA

Paradoxically, these multiple layers have not led to a finely granular design that enables complementary innovation in individual subcomponents. The different sublayers, often harking to old technology that was developed for archaic settings, aretoo numerous and complex to allow real co-design and optimization. Current standards practices have tended to generate highly prescriptive solutions for the different components of networks that suffer from the drawbacks of separate design while still proposing monolithic solutions that lack adaptability. Figure 1 shows a simplified sketch of the elements that 5G comprises.

The monolithic nature of current designs means that successes in bringing to market innovation remain piecemeal and require needlessly burdensome effort to achieve successful technology transfer. A modular approach has the potential, if embraced, to break the current logjam in 6G. While ORAN has made significant strides in modularizing management systems, physical layers remain constrained by an architecture that does not correspond or respond to current technology.

We can borrow principles from the successful approach of creating new interfaces and APIs to general units. In a manner akin to having a call to a general processing unit (GPU), we can organize systems according to units that fulfill a purpose, in a way that enables each unit to be modular, to permit incorporation of new technologies without undue obstacles. Moreover, these units can, unlike the current paradigm, be seamlessly optimized for different applications, settings and underlying technologies.

The modularization approach can allow revisiting these layers and opens the door to creating new algorithms and approaches that can dramatically improve performance. Key among the tools to improve these different modules are coding and decoding. We present here some directions of how coding and decoding can be married to this modularization in order to benefit different parts of a system and allow for greater flexibility, performance and resilience, in a way that permits innovation in the context of standards.

1. General Medium Access Unit (GMACU)

Medium access currently exemplifies the types of sedimentary layering mentioned above. ARQ, HARQ, and multiple means of reactively or proactively managing interference create an inefficient puzzle of mechanisms. Moreover, the current systems do not adapt to new technologies, such as high frequency transmissions, that benefit from different access schemes.

As an example of the above, network coding (NC) is a promising communicationsolution that can fundamentally improve the latency and resources of the current retransmission schemes such as retransmissions. These are embodied by automatic retransmission (ARQ) and hybrid (H)- ARQ in the medium access (MAC) and physical (PHY) layers, especially over lossy channels [2– 5]. The proposed GMACU will use NC as a baseline scheme to reduce/minimize retransmission while maintaining low overhead in order to meet advanced 6G requirements on latency, reliability, resource utilization, and processing costs. Scheduling algorithms and APIs can fully utilize the benefits of NC, such as flexible levels of quality of service requirements or network slice parameters, more available spectrum resources, and a greater number of users per cell.

The benefits of improving the management of losses through coding has been extensively studied: a didactic exposition can be found in [6], which covers in detail random linear network coding (RLNC). The effects of properly managing erasures can be quite dramatic. An example showing the operation and sample benefits is given in the next figure.

In medium access, emerging wireless services combine multiple network elements, such as WiFi and 5G [7,8]. Moreover, the growing number of possible frequencies available in 6G renders this combination increasingly complex. Coding across different wireless services rather than using them in a stand-alone fashion can massively reduce delay and improve reliability. This type of system permits new types of optimization possibilities that are vast in comparison with current single technology access. This rich set of design parameters enabled by coding across technologies opens the door to new optimization possibilities.

Securing these heterogeneous systems remains a challenge, particularly when different technologies may rely on hardware elements of varying provenance. In order to enable trusted communications over untrusted systems, we can consider heterogeneity as a powerful tool for security rather than as a threat. As an example, recently, a hybrid universal network coding cryptosystem (HUNCC) was put forward [9], which integrates information-theoretic and computational security to secure high-speed communication over a network in a manner that is beneficial to latency [10]. The design of the system uses a mixing method to ensure individual privacy and performs partial encryption on the mixed packets, providing the same level of computational security as full encryption. This approach significantly decreases the number of encryption and decryption operations needed at both ends by adding a mixing/demixing step. Relying on such post-quantum security can be blended into high frequency access to provide new types of physical security that are much stronger than usual mean-field arguments [11].

2. General Encoding and Decoding/Sensing Units (GEU/GDU)

Current state of the art in coding has severe limitations. These are borne out of the fact that coding and decoding have generally been viewed from a code-centric perspective. Note that traditional decoding builds codes for simplified channels, generally the worst case binary symmetric channel (BSC). Approximating that channel for traditional codes requires onerous interleaving (usually over more than a hundred bits, thus increasing latency). The first chips [12, 13] that implement Guessing Random Additive Noise Decoding (GRAND) [14,15] have provided the possibility of replacing with a single decoder the current multiple decoders, one for every type of error-correcting code and sometimes even different rates of the same code family. A soft-detection ordered reliability bits (ORB) GRAND chip achieves the lowest energy consumption of 0.76pJ/bit and lowest power consumption of 4.9mW with a throughput of 6.5Gbps and a latency of 40ns at 90MHz frequency, outperforming state-of-the-art dedicated soft-detection decoders at a target Frame Error Rate of 10⁻⁷. The ORBGRAND chip automatically adapts its performance with channel noise conditions, while supporting multiple codeword lengths.

Current codes that are purportedly capacity achieving are used quite below the capacity of channels, leading to wasting about half of the bandwidth resources, and are typically long, leading to delay and also to sharp error curves that induce frequent retransmissions for HARQ mentioned in the coding challenges. For moderate redundancy codes, which can be short and use spectrum efficiently, we can use simple IP-free codes. GRAND’s development opens up a massively larger palette of potential code-books that can be decoded with a single algorithmic instantiation, greatly reducing hardware footprint, future-proofing devices against the introduction of new codes, enabling the flexibility for each application to select the most suitable code-book, and generally opening new avenues of research for component encoding and decoding.

For longer codes with high redundancy, simple product codes, that hark back to the 1950s, can outperform 5G LDPCs while being highly flexible [16]. Interestingly, not only the BLER performance is superior. The only non-parallelizable aspect of decoding in iterative codes is basically the component decoding. LDPCs use a high number of iterations on very weak (single parity check) codes. Product codes with soft output (SO)GRAND in iterative Turbo-style decoding requires fewer iterations and are thus inherently poised to have lower latency.

By viewing non-GRAND decoders through the lens of GRAND, we can obtain SO that can render a variety of decoders suitable for iterative decoding [17, 18]. Research in a variety of product codes can open up a rich set of possibilities for lengths and rates [19].

Effectively GRAND can correct an accident of 5G, the splitting of control and data channels into two completely different types of codes, LDPCs and Polar. In contrast, 4G had both types of channels use similar decoders, with an iterative decoding for the data channels. 6G can return to this more modular and harmonious approach.

Modular design opens up a broad range of engineering possibilities, from methodologies to identify application-specific code dimensions all to be decoded with a single chip, to the use of a single small footprint to decode high-redundancy codes, in an iterative fashion, for applications that require them, and even the possibility of using cryptographic functions to simultaneously provide both data security and error correction [20]. GRAND offers the opportunity to design emerging 6G systems whilst ensuring co-existence with legacy systems within as well as in adjacent spectrum bands.

Systems opportunities that present themselves include: the real possibility of enabling accurate, low-latency error-corrected communication; the reduction of HARQ requests via receiver- side only changes by upgrading widely deployed CRCs from error detection to error correction; the provision of low-power error correction tailored for energy sensitive IoT devices; and even a fundamental re-examination of the merit of the present paradigm of substantially over- provisioning code redundancy at the expense of wasted transmission energy and latency.

3. General Interference Management Unit (GIMU)

Interference management occurs through a variety of schemes that include encoding/decoding as mentioned above, through beam steering and through multiple ways of managing interference in multi-user detection. These schemes are often only loosely tied together and work in ways that incompletely benefit from each other.

Multiple access and interference management is a prime example of where co-design of algorithms and hardware can yield significant benefits. One form of mitigation that can be incorporated in a modular way with other advances is to leverage GRAND’s innovative focus in using error correcting code-books to identify channel noise effects and invert them to correctly reveal the transmitted data. When applied to systems with multiple users, rather than decode each user individually while treating interference as noise, which results in low SNR and reduced decoding performance, GRAND can treat users in non-orthogonal multiple access (NOMA) transmissions as if they took place in augmented modulation constellations, inverting noise effects at that level, and checking for consistency by ensuring that all user data is jointly decoded correctly [22, 23]. The approach can be extended to multi-input multi-output (MIMO) systems [21]. The figure below shows the symbol error rate (SER) advantage of GRAND-AM over successive interference cancellation (SIC) in a single input/output system, as well as the advantage over V-BLAST. The system is able to manage multiple users (U users) with BPSK in a 2x2 MIMO NOMA system.

Decoding and sensing are naturally paired in current systems that consider decoding as being not just a reconstruction of the data but also a reconstruction of the environment.

4. General Transmission/Reception Unit (GTRxU)

Optimized transmission moves away from legacy prescriptive systems to make better use of spectrum resources. Current systems are highly suboptimal. It has been known for many years that optimal modulation (OM) is non-uniform in terms of constellation point placement and in terms of the probability of use of those points. The core limitation to the use of such modulations was proper reception, that can manage non-uniformity.

An error-correcting scheme can also be applied at the data-symbol level. Such error-correction requires only padding and does not depend on the transmitted symbols themselves, setting it apart from other error-correcting schemes for non-uniform inputs. This simplification allows for faster and more efficient processing while simultaneously yielding significant gains in energy efficiency often by factors of two or more. Transmission using OM yields significant improvement in error rates compared to standard QAM, achieving a ×ばつ reduction in bit error rate (BER) and a ×ばつ reduction in SER, making it ideal for applications demanding both high efficiency and robust signal transmission [24].

Conclusions

This piece provides a high-level, personal view of the benefits of modularization, particularly for the introduction of new technologies. It is meant to be neither tutorial nor comprehensive.

Without modularization, we risk ossification, entrenchment of suboptimal technologies that are often quite obsolete and, ultimately, fragility. Some of the challenges are already arising in RAN, particularly given the difficulty to compete on bespoke silicon and the highly power-hungry error-correction decoders of 5G [26]. The risk to the technology is in my view considerable.

With modularization, the onerous overhaul of successive generations could be replaced with continuous evolution of different parts of the network as the technology becomes available. Customization, differentiation, be it geographic or by verticals, can occur in a competitive, vibrant market that can welcome entrepreneurship and encourage tech transfer from academia.

It is my hope that these brief reflections will serve to foster discussion, debate and collegial critique of existing approaches, and of what will best serve the technology in the future.

References

1. M. Médard, "6G – when reliability meets performance." https://www.nokia.com/bell-labs/research/6g-networks/6g-technologies/experts-view-muriel-medard/ Accessed: 2024.
2. L. Landon, V. A. Vasudevan, J. Sung, and M. Médard, "Rethinking reliability using net- work coding: a practical 5g evaluation," in 2025 IEEE 50th Conference on Local Computer Networks (LCN), pp. 1–9, 2025.
3. E. Dias, D. Raposo, H. Esfahanizadeh, A. Cohen, T. Ferreira, M. Luís, S. Sargento, and M. Médard, "Sliding window network coding enables next generation urllc millimeter-wave networks," IEEE Networking Letters, vol. 5, no. 3, pp. 159–163, 2023.
4. A. Cohen, D. Malak, V. B. Bracha, and M. Médard, "Adaptive causal network coding with feedback," IEEE Transactions on Communications, vol. 68, no. 7, pp. 4325–4341, 2020.
5. F. Michel, A. Cohen, D. Malak, Q. De Coninck, M. Médard, and O. Bonaventure, "Flec: Enhancing quic with application-tailored reliability mechanisms," IEEE/ACM Transactions on Networking, vol. 31, no. 2, pp. 606–619, 2023.
6. M. Médard, V. A. Vasudevan, M. V. Pedersen, and K. R. Duffy, Network Coding for Engineers. John Wiley & Sons, 2025.
7. A. Cohen, G. Thiran, V. B. Bracha, and M. Médard, "Adaptive causal network coding with feedback for multipath multi-hop communications," IEEE Transactions on Communications, vol. 69, no. 2, pp. 766–785, 2021.
8. A. Cohen, H. Esfahanizadeh, B. Sousa, J. P. Vilela, M. Luís, D. Raposo, F. Michel, S. Sargento, and M. Médard, "Bringing network coding into sdn: Architectural study for meshed heterogeneous communications," IEEE Communications Magazine, vol. 59, no. 4, pp. 37–43, 2021.
9. A. Cohen, R. G. L. D’Oliveira, S. Salamatian, and M. Médard, "Network coding-based post-quantum cryptography," IEEE Journal on Selected Areas in Information Theory, vol. 2, no. 1, pp. 49–64, 2021.
10. R. G. L. D’Oliveira, A. Cohen, J. Robinson, T. Stahlbuhk, and M. Médard, "Post-quantum security for ultra-reliable low-latency heterogeneous networks," in MILCOM 2021-2021 IEEE Military Communications Conference (MILCOM), pp. 933–938, IEEE, 2021.
11. C.-Y. Yeh, A. Cohen, R. G. L. D’Oliveira, M. Médard, D. M. Mittleman, and E. W. Knightly, "Securing angularly dispersive terahertz links with coding," IEEE Transactions on Information Forensics and Security, vol. 18, pp. 3546–3560, 2023.
12. A. Riaz et al., "Multi-code multi-rate universal maximum likelihood decoder using GRAND," in ESSCIRC 2021 - IEEE 47th European Solid State Circuits Conference (ESS- CIRC), (Grenoble, France), 2021.
13. A. Riaz et al., "A sub-0.8 pJ/bit, 16.3 gbps/mmˆ2 universal soft-detection decoder using orbgrand in 40 nm CMOS," in IEEE International Solid-State Circuits Conference (ISSCC), 2023.
14. K. R. Duffy, J. Li, and M. Médard, "Capacity-achieving guessing random additive noise decoding," IEEE Transactions on Information Theory, vol. 65, pp. 4023–4040, July 2019.
15. K. R. Duffy, W. An, and M. Médard, "Ordered reliability bits guessing random additive noise decoding," IEEE Transactions on Signal Processing, vol. 70, pp. 4528–4542, 2022.
16. P. Yuan, M. Médard, K. Galligan, and K. Duffy, "Soft-output (SO) GRAND and iterative decoding to outperform LDPC codes," IEEE Transactions on Wireless Communications, January 2025.
17. P. Yuan, K. Duffy, and M. Médard, "Soft-output successive cancellation list decoding," IEEE Transactions on Information Theory, vol. 71, pp. 1007–1017, February 2025.
18. K. Duffy, P. Yuan, J. Griffin, and M. Médard, "Soft-output guessing codeword decoding," IEEE Communications Letters, vol. 29, February 2025.
19. S. Khalifeh, K. Duffy, and M. Médard, "Turbo product decoding of cubic tensor codes," in Conference on Information Sciences and Systems (CISS), March 2025. Best Paper Award.
20. A. Cohen, R. G. L. D’Oliveira, K. R. Duffy, and M. Médard, "Partial encryption after encoding for security and reliability in data systems," in 2022 IEEE International Symposium on Information Theory (ISIT), pp. 1779–1784, IEEE, 2022.
21. K. Yang, M. Médard, and K. Duffy, "GRAND-AM in uplink MIMO NOMA systems," in Conference on Information Sciences and Systems (CISS), March 2025.
22. K. Yang, M. Médard, and K. Duffy, "Nonorthogonal multiple access with guessing random additive noise decoding-aided macrosymbol (GRAND-AM)," IEEE Internet of Things Journal, vol. 11, pp. 28036–28049, June 2024.
23. A. Solomon, K. Duffy, and M. Médard, "Managing noise and interference separately - multiple access channel decoding using soft GRAND," in IEEE International Symposium on Information Theory (ISIT), 2021.
24. T. Zirtiloglu, A. Tan, B. Ozaydin, K. Duffy, M. Médard, and R. Yazicigil, "Fully integrated optimal modulation bits-to-RF digital transmitter using time-interleaved multi-subharmonic-switching DPA," in IEEE Radio Frequency Integrated Circuits Symposium (RFIC), 2025.
25. B. Ozaydin, M. Médard, and K. Duffy, "GRAND-assisted demodulation," IEEE Journal on Selected Areas in Communications: Next Generation Advanced Transceiver Technologies, November 2024.
26. I. Morris, "What Intel crisis means for Samsung and Nokia in 5G." https://www.lightreading.com/5g/what-intel-crisis-means-for-samsung-and-nokia-in-5g. Accessed: 2025.