5G Network Implementation

With the current impact of cell network outages across almost all carriers in the US, it's a good time to talk about the future; actually, it's not even about the future, it's the present. Several years ago I started talking about having mobile robotics (air, ground and maritime robotics, like drones, rovers and submergible devices) be part of a mobile adhoc network or MANET. One example is a private mesh network, like Silvus Technologies provides. These communications solutions for high bandwidth video, C2, health and telemetry data are absolutely needed in today's environment and allow for a very flexible set-up and coverage; from a local incident scene, to a much larger area coverage, to entire cities or counties being covered. Why the need? While we in the drone industry originally focused on getting drones connected to a cell network, we quickly realized the single point of failure; the cell network infrastructure. Natural disasters, as well as manmade disasters, can impact these networks dramatically. An earthquake, hurricane, a solar storm, or a cyberattack, can take down these public networks for hours to days. And that includes public safety dedicated solutions like FirstNet or Frontline, during times when coms and data push is absolutely needed. Over the past couple of years we have seen the rise of mobile robotics deployments within private networks. While the defense side has done this approach for years, the public safety sector is still new to this concept. Some solutions integrate with a variety of antennas, amplifiers and ground stations, offer low latency, high data rates (up to 100+Mpbs), 256-bit AES encryptions and allow for a very flexible and scalable mobile ad-hoc mesh network solution. And most importantly - independence from a public network system. And now imagine you have multiple devices operating; a helicopter, a drone, a ground robotic, together with individuals on the ground, all connected and all tied into a geospatial information platform, like ATAK/TAK. Each connected device can become a node and extend the range. This is what I am calling building the Tech/Tac Bubble. This is not just the future, this is already happening with a handful of agencies across the US It's time to start thinking about alternative communication solutions and mobile robotics are an important part of leading the way. #UAV #UAS #UGV #Drones #network #MANET #Meshnetwork #publicsafety

Frequency Escalation in UAV Systems – Transmissions in the 7.5–12 GHz Band Recent observations indicate a clear upward shift in the radio spectrum used by unmanned aerial systems (UAS). Traditional ranges for command and video links — 300 MHz to 7.2 GHz — are now heavily saturated. Consequently, more UAVs are operating within the 7.5–12 GHz band, entering the centimeter-wave (SHF) domain rarely used by small and medium-class drones. Field reports confirm analog video transmitters above 8 GHz, marking a significant departure from the standard 2.4 GHz and 5.8 GHz bands. Operating higher enables avoidance of interference and greater data throughput, especially for HD and 4K video with minimal latency. This, however, demands high RF precision and antenna stability, as even minor detuning degrades link performance. Frequencies above 7 GHz mean shorter wavelengths, faster attenuation, limited obstacle penetration, and strict line-of-sight requirements. Maintaining stable connections requires high-gain directional antennas, increased transmitter power, or airborne relay UAVs to sustain long-range links despite terrain masking. Operation in the 8–12 GHz range allows wider bandwidth and lower latency but requires advanced RF filtering, thermal stabilization, and high-linearity amplification (LNA/PA). This raises system complexity while reducing detectability. Most current detection and counter-UAS (C-UAS) systems cover up to ~7 GHz. Thus, new UAVs may operate beyond detection. Analog modulation at these frequencies generates non-standard spectral signatures not recognized by common RF classification algorithms. To adapt, infrastructures must expand spectrum monitoring to at least 12 GHz, update RF signature libraries, upgrade analyzer firmware, and test jamming effectiveness in the 8–12 GHz range. The ongoing upward shift in UAV frequencies marks a new phase in unmanned architecture, emphasizing adaptability, dynamic channel allocation, and resilience in contested electromagnetic environments. The spectrum itself has become a battlefield — one where superiority depends on intelligence, agility, and precise spectrum management.

📡 C-Band, E-Band, and K-Band 🔹 C-Band • Frequency Range: ~4 GHz to 8 GHz • Use in Telecom: Satellite communication, microwave backhaul (long distance). • Advantages: • Good coverage over long distances. • Less rain fading compared to higher bands. • Disadvantages: • Requires bigger antennas. • Lower capacity compared to E/K band. ⸻ 🔹 E-Band • Frequency Range: ~71 GHz to 86 GHz • Use in Telecom: High-capacity microwave backhaul, especially in 4G/5G networks. • Advantages: • Very high data rates (multi-Gbps). • Smaller antennas. • Disadvantages: • Sensitive to rain fading. • Short distance (typically < 3–5 km). ⸻ 🔹 K-Band • Frequency Range: ~18 GHz to 27 GHz • Use in Telecom: Satellite communication (Ku/K band), radar, and medium-distance microwave links. • Advantages: • Higher capacity than C-band. • Antennas are smaller. • Disadvantages: • More rain attenuation than C-band. • Coverage distance shorter. 🎯 Interview Answer (short): “C-band (4–8 GHz) is used for long-distance links with less rain fade but lower capacity. K-band (18–27 GHz) supports medium-distance links with higher capacity but more rain attenuation. E-band (71–86 GHz) is used in modern 5G backhaul, providing multi-Gbps capacity but only over short distances due to heavy rain fa

France is the first in Europe to launch a public safety network that combines a sovereign, state-controlled core with priority roaming across multiple independent public RANs. It is a model of resilience, distinct from the UK’s ESN and the US’ FirstNet, with unique operator diversity, and it embeds a capability to stand up 4G coverage nationally in less than six hours via deployable 4G/5G sites with complete energy autonomy and satellite backhaul. The mission-critical MCX network (known as Réseau Radio du Futur, or RRF) is replacing the legacy Tetrapol system used in France and has developed under a top-down, state-led approach with the (prudent) selection of domestic vendors/system integrators (Airbus and Capgemini) to preserve technological sovereignty and national security. While resilient, the legacy Tetrapol network was hindered by fragmentation, with bespoke systems used by different agencies (problematic when a multi-agency response was required), and lacked modern capabilities like video due to extremely low throughput capability of the 2G-based infrastructure. The agency tasked with operating the new network, ACMOSS, announced this week that Bouygues Telecom's 4G/5G RAN is now live for the RRF platform with priority access and pre-emption, adding a second competing independent mobile network alongside Orange and removing the single-MNO dependency observed in other national public safety networks. The RRF system also provides last resort national roaming capability (although not priority access) to Free and SFR's infrastructure, further enhancing redundancy at the air interface level across different spectrum and site footprints. Based on a full-MVNO architecture anchored in a state-run core, it provides subscription-based priority and pre-emption (QPP) on the host networks while keeping authentication, security, MCX and policy control in the RRF core. This model is similar to MOCN in outcomes but not MOCN technically (yet), since there is no direct RAN sharing, instead, home-routed roaming (S8HR-style) is used, which may add some latency but preserves sovereign control over services and policy. First responder devices use the ATRIA app to pick the best network (can be force-switched if needed). On top of the multi-RAN and sovereign core setup, the RRF stack also includes layered deployables (similar to FirstNet) and direct mode capability for off-network continuity in the most extreme scenarios. Lightweight vehicular kits have been designed to create local "Wi-Fi bubbles" quickly for on-scene operations or in-building ingress with 4G/5G modems, rapid-response assets can project 4G coverage nationally within hours using energy-autonomous units with satellite backhaul and heavier deployable eNB/gNB solutions can sustain wide-area coverage for days. RRF handsets are also paired with a direct-mode accessory (i.e.,DMR) to maintain talk-group comms when there is no cellular service.

Zero Copy Data Transfer in HPC: A common technique for loading data in high performance applications is called “zero copy” because, well, it doesn’t require a copy. But what does that mean, and why is it useful? As I harp on in many of my posts, data movement is typically one of the largest bottlenecks and biggest challenges in high performance computing today. If we think about a 405B parameter LLM, we are transferring around, at a minimum, 405GB of data in memory. But this is virtually nothing when compared to the petabytes of data required to train that model. Traditional data transfer methods involve multiple copying of data between user space and kernel space, leading to increased CPU usage and reduced throughput. Let’s dive deeper: Problems with traditional data transfer: In a conventional data transfer operation, say from disk to a network interface, the data typically goes through multiple stages: - Reading from disk into kernel buffer - Copy from kernel buffer to user space - transform and copy back to kernel before network send - transmitted to network interface for sending Each requires a copy, requiring cpu cycles and memory bandwidth ultimately becoming rate limiting for large data. How Zero Copy Works: Zero Copy eliminates redundant data copies by using system-level techniques that allow data to be transferred directly between kernel space and the target destination without intermediary copies. Several Zero Copy techniques are implemented in modern operating systems: - Memory Mapping (mmap): mmap allows files to be mapped directly into the address space of a process. This means that the file contents can be accessed as if they were in memory, reducing the need for copying between kernel and user space. - Sendfile(): In networked applications, the sendfile() system call enables data to be sent directly from a file descriptor (such as a file on disk) to a socket, bypassing user space entirely. - Direct I/O: Direct I/O bypasses the kernel’s buffering mechanisms, allowing data to be read or written directly to and from disk. - DMA (Direct Memory Access): hardware-level technique where data is transferred directly between the memory and a device without CPU intervention. Ultimately, zero copy provides reduced CPU utilization, lower latency access, increased throughput, and more efficient memory usage. Several technologies exist that leverage zero copy architecture directly, such as GPU Direct Storage by NVIDIA, RDMA over Converged Ethernet, and even Network Filesystems. Diving into understanding this will help you better understand how to efficiently move data in your HPC applications. If you like my content, feel free to follow or connect! #softwareengineering #hpc

A sluggish API isn't just a technical hiccup – it's the difference between retaining and losing users to competitors. Let me share some battle-tested strategies that have helped many  achieve 10x performance improvements: 1. 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗖𝗮𝗰𝗵𝗶𝗻𝗴 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝘆 Not just any caching – but strategic implementation. Think Redis or Memcached for frequently accessed data. The key is identifying what to cache and for how long. We've seen response times drop from seconds to milliseconds by implementing smart cache invalidation patterns and cache-aside strategies. 2. 𝗦𝗺𝗮𝗿𝘁 𝗣𝗮𝗴𝗶𝗻𝗮𝘁𝗶𝗼𝗻 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 Large datasets need careful handling. Whether you're using cursor-based or offset pagination, the secret lies in optimizing page sizes and implementing infinite scroll efficiently. Pro tip: Always include total count and metadata in your pagination response for better frontend handling. 3. 𝗝𝗦𝗢𝗡 𝗦𝗲𝗿𝗶𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻 This is often overlooked, but crucial. Using efficient serializers (like MessagePack or Protocol Buffers as alternatives), removing unnecessary fields, and implementing partial response patterns can significantly reduce payload size. I've seen API response sizes shrink by 60% through careful serialization optimization. 4. 𝗧𝗵𝗲 𝗡+𝟭 𝗤𝘂𝗲𝗿𝘆 𝗞𝗶𝗹𝗹𝗲𝗿 This is the silent performance killer in many APIs. Using eager loading, implementing GraphQL for flexible data fetching, or utilizing batch loading techniques (like DataLoader pattern) can transform your API's database interaction patterns. 5. 𝗖𝗼𝗺𝗽𝗿𝗲𝘀𝘀𝗶𝗼𝗻 𝗧𝗲𝗰𝗵𝗻𝗶𝗾𝘂𝗲𝘀 GZIP or Brotli compression isn't just about smaller payloads – it's about finding the right balance between CPU usage and transfer size. Modern compression algorithms can reduce payload size by up to 70% with minimal CPU overhead. 6. 𝗖𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗼𝗻 𝗣𝗼𝗼𝗹 A well-configured connection pool is your API's best friend. Whether it's database connections or HTTP clients, maintaining an optimal pool size based on your infrastructure capabilities can prevent connection bottlenecks and reduce latency spikes. 7. 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗟𝗼𝗮𝗱 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 Beyond simple round-robin – implement adaptive load balancing that considers server health, current load, and geographical proximity. Tools like Kubernetes horizontal pod autoscaling can help automatically adjust resources based on real-time demand. In my experience, implementing these techniques reduces average response times from 800ms to under 100ms and helps handle 10x more traffic with the same infrastructure. Which of these techniques made the most significant impact on your API optimization journey?

5G Registration Procedure in Network Slicing Ever wondered how network slicing influences the 5G registration process? Let’s walk through how your device (UE) not only connects but also negotiates which slice it belongs to. 🔹 Step 1: UE Sends Requested NSSAI When UE initiates registration, it includes Single NSSAI essentially saying: “I want access to a specific 5G slice.” Example parameters: SST (Slice/Service Type): 1 = eMBB, 2 = URLLC, 3 = mMTC SD (Slice Differentiator): Distinguishes between multiple slices of the same type Supported AMFs: Lists which AMFs the UE can connect to Default AMF: UE’s preferred AMF for registration Think of it as the UE handing the network a “slice preference card” right at registration. Step 2: AMF Selection Based on s-NSSAI This is where slicing begins to shape control-plane behavior. 1️⃣ The gNB forwards the registration request to a default or UE-specified AMF. 2️⃣ The AMF queries the NSSF (Network Slice Selection Function) to check: Is the requested slice available in this PLMN? Is this AMF authorized to serve it? If not, the AMF can: Redirect the UE to a different AMF Reject the requested slice Modify the Allowed and Configured NSSAI lists Slicing directly influences how and where the UE’s control plane is anchored. Step 3: Security Setup Before Slice Authorization Even before slice approval, standard NAS security procedures run first — including identity verification and authentication. Only after successful authentication can the UE proceed to slice-specific authorization. Authenticate first. Authorize slices next. Step 4: Slice Authorization & Context Setup The AMF now finalizes the slice configuration by sending the Initial Context Setup Request to the gNB, containing: Allowed NSSAI: Slices the UE is authorized to use Configured NSSAI: Active subset for this registration Rejected NSSAI: Slices not supported or permitted NSSAI Inclusion Mode: Rules for signaling slice info in later procedures Example: UE requests eMBB, URLLC, and a private slice. The NSSF checks the configuration and returns: ✅ eMBB & URLLC allowed ❌ Private slice rejected (not deployed in this PLMN) The UE learns this via the Registration Accept message. ✅ Step 5: UE Completes Registration with Slice Awareness After receiving Registration Accept and completing RRC Reconfiguration, the UE now knows: Which slices it can use Which were rejected Which AMF it’s anchored to It’s now ready to initiate PDU Session Establishment per slice with appropriate QoS and user plane configuration. 🎓 Want to master 5G slicing procedures end-to-end? Enroll in our course with topic “5G Core Signaling & Network Slicing" 👉 https://lnkd.in/e3S6B4aW #5G #NetworkSlicing #5GCore #NSSAI #AMF #NSSF #TelecomTraining #5GArchitecture #RAN #Telecommunications

Following the strong engagement on my recent #Echo update, I want to share a more technical look at #Tabua — Trans Pacific Networks’ #subsea cable system targeting Ready for Service in 2026. Tabua introduces a new #Australia ⇄ #UnitedStates optical corridor, expands #Oceania connectivity, and integrates directly into major metro PoPs in #Sydney and #LosAngeles, enhancing network diversity and interconnection options across the #Pacific. Engineering Characteristics - • Supplier: SubCom • Total Fiber Pairs: 16 • #TPN Ownership: 1 full fiber pair (end-to-end) • Topology (landing points):   • Australia   • #Fiji   • #Hawaii   • U.S. West Coast • Architecture: Repeatered long-haul system with multiple landing branches Engineering Significance - • Provides a new long-haul optical path between Australia and the United States • Adds diversity to existing Australia ⇄ U.S. routes • Introduces additional subsea interconnection options in Fiji and Hawaii • Supports scalable wavelength capacity for #carriers, #cloudoperators, #globalenterprises, #AI • Enhances network resilience and routing flexibility across Oceania and into North America Tabua strengthens the subsea infrastructure ecosystem by bringing a distinct and complementary path to other Pacific cable systems, increasing overall network health and diversity. PoP-to-PoP Architecture for Tabua - Below is a high-level view of TPN’s PoP strategy aligned with Tabua’s subsea landing infrastructure and backhaul topology. 🇦🇺 Sydney PoP — #Equinix • Located in Equinix Sydney, one of Australia’s primary cloud, carrier, and digital infrastructure hubs • Provides direct interconnection into major Australian cloud regions and content networks 🇺🇸 Los Angeles PoP — Equinix • Located in Equinix Los Angeles, a major west-coast interconnection market • Provides direct access to cloud platforms, media networks, and global backbone carriers 🇺🇸 San Jose PoP — Equinix (Extended U.S. Access) • Located in Equinix San Jose, a key Silicon Valley connectivity region supporting cloud, AI, and hyperscale compute • Provides an additional northern California PoP option for Tabua customers TPN’s Product Offering on Tabua - • #Lease & #IRU • Ethernet Waves: #10G, #100G, #400G • #Spectrum Our flexible product suite enables customers to design solutions ranging from dedicated, high-capacity wavelengths to managed spectrum services across the Australia ⇄ U.S. corridor. Tabua is progressing toward Ready for Service in 2026, and TPN is proud to contribute to a more diverse, resilient, and scalable subsea infrastructure ecosystem across the Pacific. If you’d like to learn more about our capabilities, feel free to connect with me or anyone from the Trans Pacific Networks team: Aaron Knapik, @Mira Ivanac, Lee Kerridge, Robin Pula, Gavin Tully, Howard Kidorf, Philip deGuzman, Austin Shields, Jonathan Javier, David Finch #Subsea #SubmarineCables #AustraliaToUSA #Oceania #NetworkPlanning #AI #PTC2026

Designing an AI System That Doesn’t Collapse Under Latency Spikes A single user query passes through multiple stages — tokenization → batching → GPU scheduling → model execution → post-processing → response assembly. Now picture this: A few heavy prompts take 5× longer than average. Your batching layer waits to fill the “perfect batch.” Meanwhile, the queue grows. Requests start timing out. Retries stack up. That’s when you realize: You’re not running out of compute. You’re running out of control. Here’s how you design for resilience instead of collapse 👇 1️⃣ Bounded Queues Never let latency scale linearly with load. Bound your input queues and shed load proactively — either by dropping excess requests or serving degraded responses. Unbounded queues are silent killers — they delay backpressure, causing cascading timeouts. Think of it like circuit breakers for inference — graceful denial is better than system-wide collapse. 2️⃣ Adaptive Batching Static batch sizes look great in benchmarks and terrible in production. Instead, make batch sizes dynamic — continuously tuned based on GPU occupancy, queue length, and recent tail latency percentiles (P95/P99). At low load, batch small for lower latency. At high load, batch large for throughput — but with strict timeouts. The goal is elasticity without unpredictability. 3️⃣ Token-Aware Scheduling Batching by request count is naive. In LLM workloads, token length determines cost. A single 10,000-token prompt can stall 15 smaller ones if batched together. Token-aware schedulers measure total token budget per batch and allocate GPU time accordingly. This ensures fairness and consistent latency curves even under mixed workloads. 4️⃣ Partial Caching Most engineers cache final model outputs. That helps little. What actually saves time is pre- and post-compute caching — tokenized inputs, embeddings, and prompt templates. These are deterministic and cheap to reuse, shaving milliseconds off critical paths. Combine that with vector cache lookups to skip redundant reasoning altogether. 5️⃣ Deadline-First Scheduling In multi-tenant inference systems, not all requests are equal. Prioritize requests based on expected completion deadlines instead of FIFO order. This minimizes tail latency and improves QoS across traffic tiers. It’s the same principle airlines use — business class boards first, but everyone still gets there. This is where systems engineering meets AI infrastructure. Because LLM inference at scale isn’t just about throughput — it’s about temporal predictability. Inside my Advanced System Design Cohort, we go deep into these challenges — how to design AI systems that don’t just scale, but stay stable under load. If you’ve been leading distributed systems or AI infra and want to sharpen your architectural depth, there’s a link to a form in the comments — apply, and we’ll check if you’re a great fit.

When working with 𝗟𝗟𝗠𝘀, most discussions revolve around improving 𝗺𝗼𝗱𝗲𝗹 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆, but there’s another equally critical challenge: 𝗹𝗮𝘁𝗲𝗻𝗰𝘆. Unlike traditional systems, these models require careful orchestration of multiple stages, from processing prompts to delivering output, each with its own unique bottlenecks. Here’s a 5-step process to minimize latency effectively:  1️⃣ 𝗣𝗿𝗼𝗺𝗽𝘁 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: Optimize by caching repetitive prompts and running auxiliary tasks (e.g., safety checks) in parallel.  2️⃣ 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: Summarize and cache context, especially in multimodal systems. 𝘌𝘹𝘢𝘮𝘱𝘭𝘦: 𝘐𝘯 𝘥𝘰𝘤𝘶𝘮𝘦𝘯𝘵 𝘴𝘶𝘮𝘮𝘢𝘳𝘪𝘻𝘦𝘳𝘴, 𝘤𝘢𝘤𝘩𝘪𝘯𝘨 𝘦𝘹𝘵𝘳𝘢𝘤𝘵𝘦𝘥 𝘵𝘦𝘹𝘵 𝘦𝘮𝘣𝘦𝘥𝘥𝘪𝘯𝘨𝘴 𝘴𝘪𝘨𝘯𝘪𝘧𝘪𝘤𝘢𝘯𝘵𝘭𝘺 𝘳𝘦𝘥𝘶𝘤𝘦𝘴 𝘭𝘢𝘵𝘦𝘯𝘤𝘺 𝘥𝘶𝘳𝘪𝘯𝘨 𝘪𝘯𝘧𝘦𝘳𝘦𝘯𝘤𝘦.  3️⃣ 𝗠𝗼𝗱𝗲𝗹 𝗥𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀: Avoid cold-boot delays by preloading models or periodically waking them up in resource-constrained environments.  4️⃣ 𝗠𝗼𝗱𝗲𝗹 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: Focus on metrics like 𝗧𝗶𝗺𝗲 𝘁𝗼 𝗙𝗶𝗿𝘀𝘁 𝗧𝗼𝗸𝗲𝗻 (𝗧𝗧𝗙𝗧) and 𝗜𝗻𝘁𝗲𝗿-𝗧𝗼𝗸𝗲𝗻 𝗟𝗮𝘁𝗲𝗻𝗰𝘆 (𝗜𝗧𝗟). Techniques like 𝘁𝗼𝗸𝗲𝗻 𝘀𝘁𝗿𝗲𝗮𝗺𝗶𝗻𝗴 and 𝗾𝘂𝗮𝗻𝘁𝗶𝘇𝗮𝘁𝗶𝗼𝗻 can make a big difference.  5️⃣ 𝗢𝘂𝘁𝗽𝘂𝘁 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀: Stream responses in real-time and optimize guardrails to improve speed without sacrificing quality. It’s ideal to think about latency optimization upfront, avoiding the burden of tech debt or scrambling through 'code yellow' fire drills closer to launch. Addressing it systematically can significantly elevate the performance and usability of LLM-powered applications. #AI #LLM #MachineLearning #Latency #GenerativeAI