Architecture review: BLE-based indoor navigation system for visually impaired ra

[nachiket26]

I'm designing a final-year major project: a BLE-based assistive navigation system for visually impaired people in railway stations.

The basic idea is to place low-power BLE beacons at key navigation points—stairs, lifts, turns, platform entrances, exits, etc. A wearable ESP32 detects these beacons and provides haptic/audio cues such as "turn left", "lift ahead", or "take the staircase."

For the first prototype, the system will use predefined routes: the user enters the station, receives the relevant instructions from the beacons, and subsequent beacons act as checkpoints to confirm that they're following the route.

The more advanced version I'm considering is:

• A station is represented as a graph, where beacons are nodes and only physically walkable connections are edges.
• The wearable can obtain the station graph and run Dijkstra/A* locally to calculate a route to a requested destination.
• Beacons then act primarily as location checkpoints, allowing the wearable to detect whether the user is following the planned route and re-route if necessary.
• The graph can contain constraints such as closed lifts/stairs, inaccessible paths, platform boundaries, etc.
• A central station-management system would maintain the master graph and distribute updated versions to the infrastructure.
• I'm also exploring how battery-powered beacons could receive updates without requiring wired connectivity or internet on every beacon, potentially using a gateway/mesh-style approach.

Some questions I'm currently trying to solve:

1. How should the BLE beacon hardware be designed for low cost and long battery life?
2. What is the best way to distribute/update the station graph to many offline beacons?
3. How reliable is BLE/RSSI for checkpoint detection in a crowded railway environment?
4. How should the system handle failed/missing beacons and multiple nearby beacons?
5. Is BLE Mesh or another relay architecture actually appropriate here?
6. What important real-world failure cases am I overlooking?
7. Does this architecture make sense at all, or am I overengineering the problem?

I'm deliberately sharing this before finalizing the architecture. I'd really appreciate criticism from people experienced in BLE/RF, embedded systems, networking, IoT, accessibility, or indoor navigation.

I'm much more interested in what is wrong with this approach and what I haven't considered than in validation.

[nali]

Indoor navigation is the holy grail for blind & visually impaired and it's been looked at for a loooong time.

E.g. I remember attending a seminar hosted at Microsoft Research on this, over a decade ago:
https://newatlas.com/microsoft-bone-conduction-headset-for-blindess/34622/
Although that was generic location based, one of the topics under discussion was indoor navigation. These guys https://www.pointr.tech/solutions/location-based-services/indoor-navigation were part of that discussion.

Just using BLE is impractical at places like public transit. Lots of walls and metalwork for signals to reflect and at places like bus stations entrances/exits can be literally a few metres apart

[rf-fil]

1. How should the BLE beacon hardware be designed for low cost and long battery life?
2. What is the best way to distribute/update the station graph to many offline beacons?
3. How reliable is BLE/RSSI for checkpoint detection in a crowded railway environment?
4. How should the system handle failed/missing beacons and multiple nearby beacons?
5. Is BLE Mesh or another relay architecture actually appropriate here?
6. What important real-world failure cases am I overlooking?
7. Does this architecture make sense at all, or am I overengineering the problem?

You're massively underestimating the difficulty of this problem. BLE does not work for navigation. There are issues with radio propagation, hardware heterogeneity, deployment and maintenance.

• RSSI is a poor distance proxy. Range estimation relies on a log-distance path loss model whose exponent varies from roughly 1.6 to 6 depending on the environment, and is not constant within a single building. Errors in the exponent produce huge distance errors.
• Multipath fading. Reflections off metal, glass, concrete, etc., cause constructive and destructive interference. RSSI at a fixed point can vary ±20 dB or more.
• Human body attenuation. 2.4 GHz is strongly absorbed by water. A body between beacon and receiver costs about 5 to 15 dB. Accuracy degrades at peak occupancy, when navigation is most needed.
• Co-channel interference. The 2.4 GHz ISM band is shared with Wi-Fi, microwave ovens, and other traffic.
• Floor leakage. Signal propagates vertically through slabs, stairwells, and atria, making floor discrimination unreliable. Vertical errors are qualitatively worse than horizontal ones.
• Antenna orientation and occlusion. Wearable in a pocket, bag, or held at an angle changes received power substantially. Neither beacon nor wearable antennas are isotropic.
• Advertising interval latency. Intervals of 100 ms - 1 s combined with filtering means the estimate lags a walking user by several metres.
• Physical disturbance. Beacons are moved, removed, painted over, or occluded by inventory and signage.
• Radio map staleness. Fingerprinting approaches require a labour-intensive site survey that is invalidated by layout changes or new fixtures. Re-surveying is rarely budgeted.
• Geometric dilution of precision. Trilateration requires three or more beacons in range with favourable geometry. Corridors force near-collinear arrangements, producing large error ellipses along the axis of travel.

Realistic BLE accuracy is within maybe 5m in about 50% of the time, if I'm being optimistic and much worse the rest of the time. A 5-meter error places the user in a different room or on the wrong side of a wall.

[xvr]

RSSI isn't suitable for estimating distance (just as rt-fil said).

Bluetooth Channel Sounding is a much better option: https://www.silabs.com/wireless/bluetooth/channel-sounding

[gerber]

Relevant link from u-blox that have done some work on bluetooth indoor positioning
https://www.u-blox.com/en/technologies/bluetooth-indoor-positioning

[xvr]

Relevant link from u-blox that have done some work on bluetooth indoor positioning
https://www.u-blox.com/en/technologies/bluetooth-indoor-positioning

Yes, this is a possible solution. But it relies on "direction finding," which requires an antenna array (at least two antennas on the beacon or receiver).

[nachiket26]

I also got suggestions to use UWB (Ultra Wide Band). but for a college project it will get expensive to install UWB modules.

One thing I am confused about - Should I prioritize Location accuracy or Navigation/Routing. 

• I was thinking I don't need to accurately locate the user, instead the beacons are just checkpoints to the Routing.
• Now, I haven't figured out how am I going to route the user, hypothetically I can say that the station map with beacons as nodes is stored somewhere and it will be transmitted to the user wearable device which will then route the user to its destination. Meanwhile these intermediary beacons act as checkpoint to make sure the user is on correct path.
• So I am wondering if precise location of the user is even necessary? If beacons are just checkpoints

I would appreciate if anyone can clarify my doubt. I will learn something from it

[xvr]

So I am wondering if precise location of the user is even necessary?

Yes, unless you want to direct the user toward the wall instead of the hallway. The difference can be less than 1 meter.

[nctnico]

I also got suggestions to use UWB (Ultra Wide Band). but for a college project it will get expensive to install UWB modules.

UWB is indeed the technology to use for indoor positioning. But you'll also need very accurate time synchronisation in order to achieve accuracy within a meter. Low double digit picoseconds between nodes.

A better route is probably a system using cameras and AI. Make a person wear a marker and triangulate the position using cameras. There are probably cameras in a railway station already and the system can also serve to track out-of-character individuals which may need some attention.

[nali]

So I am wondering if precise location of the user is even necessary? If beacons are just checkpoints
I would appreciate if anyone can clarify my doubt. I will learn something from it

Put yourself in the shoes of the intended user. You cannot see. Headphones say "turn left" so you turn left. If you don't have your location down to a couple of metres or less (arc of a stick) you might be walking into a wall or even off the end of a railway platform.

If your goal is journey auditing/ tracking maybe OK. Navigation? Nah.

[coromonadalix]

ok  suggestion here   why not lidar based ?

i woul dnever use BLE  or any rf sensitive devices


lidar is laser

[nachiket26]

Do u mean to use Lidar sensor on the wearable that the user carry?
Lidar seems good for obstacle detection. Idk if navigation is possible using it.

Also considering the rush on railway station we have here in India, Lidar will false alarm everyone as an obstacle