Do we have proper enclosures with enough antenna openings? Many of the previous cases lacked sufficient cutouts to accommodate all 5G and WiFi antennas. For example, modules usually needs 4 antennas, and new one like the FM190W-GL or RM551 require 7 antennas, and mmWave modules may need up to 12. WiFi 7 also typically requires much more antennas (bpi’s be19000 says 14 Antenna, so no more sma hole for 5g). If the board/wifi module+5g module supports full antenna configurations but the enclosure imposes limitations, it becomes a design bottleneck — and far from ideal.
Additionally, is it really advisable to connect all modules via CPU-driven drivers, especially when multiple vendors are involved (e.g., Fibocom, Quectel, Meig, TD-Tech)? This can introduce considerable performance overhead and stability risks. And, I’ve had the BPI-R4 for over a year, and there still isn’t a stable, production-grade firmware available.
Some other 5G CPE/router vendors seem to be taking a better approach by using 2.5G/1G PHYs — which I believe is the optimal solution. In this setup, modern cellular modules (including quectel/fibocom/td-tech all support this way nativly) operate in PCIe RC mode and use interfaces like RTL8111/8125 (and newer modules such as FM190W also support attaching AQC 5GbE/10GbE NICs), connecting to the router through standard Ethernet. This design avoids the need for vendor-specific drivers (such as rmnet, QMI, RNDIS, ECM, or NCM), many of which are buggy, rely heavily on vendor patches, and disable hardware acceleration — leading to increased CPU load.
If all modules were connected via Ethernet-like interfaces and routed to the 7988 through a switch chip (with VLAN segmentation), it would save PCIe lane/USB bus bandwidth & routing complexity and significantly reduce CPU overhead. In effect, module connections would behave just like standard Ethernet or SFP WAN interfaces, fully leveraging hardware-accelerated forwarding (and most important, it doesn’t require vendor driver). This architecture would also enhance scalability by removing PCIe/USB bottlenecks.
By the way, it’s still a very promising board — is it currently available anywhere (e.g., Taobao)?