Debugging Hardware in Production: How We Solved Dual-Barcode Interleaving on Industrial Handheld Scanners published

text
Full stream:   4 6 5 4 8 6 5 5 1 8 3 5 8 1 4 3 7 8 1 4 8 7 6 1 4 8 2 6
Odd index:     4   5   8   5   1   3   8   4   7   1   8   6   4   2   -> 465851384718642 (15 digits: The real waybill!)
Even index:      6   4   6   5   8   5   1   3   8   4   7   1   8   6 -> 64658513847186 (13 digits: Truncated duplicate scan)



The scan engine had interleaved two concurrent decodes character-by-character into a single rapid keyboard burst!





2. Why Naive Fixes Failed



Before writing algorithms, team discussions usually produce three common suggestions:








"Configure the scanner hardware to read only one barcode at a time."

Reality: In contract logistics, warehouses operate dozens of device models from different manufacturers (Urovo, Zebra, Chainway, Honeywell) running custom Android ROMs. Relying on manual field hardware configuration across hundreds of workers is fragile.








"Hardcode courier prefix and length rules."

Reality: Couriers frequently update numbering schemes, add sub-carriers, or introduce new routing codes. A hardcoded courier dictionary creates maintenance debt and breaks whenever a new parcel format is introduced.








"De-bounce or throttle keystrokes in onChange."

Reality: Running aggressive de-duplication on every keystroke prefix breaks normal human keyboard entry. If an operator types a genuine number with consecutive digits, an over-eager filter can silently truncate real data.









3. The Solution: Carrier-Agnostic De-Interleaving



We solved this at the software input layer by isolating the completion trigger (scanner Enter or Blur event) and applying a deterministic de-interleaving algorithm.




Here is a simplified TypeScript implementation of how we extract the true base tracking number:



/**
 * Detects and repairs interleaved concurrent barcode scans.
 * Recovers the base number when two rapid scans interlace alternating characters.
 */
export function splitInterleavedScan(input: string): string {
  const clean = input.trim();

  // Scans below minimum tracking length are untouched
  if (clean.length < 16) {
    return clean;
  }

  // Decompose into odd and even character streams
  let streamA = "";
  let streamB = "";

  for (let i = 0; i < clean.length; i++) {
    if (i % 2 === 0) {
      streamA += clean[i];
    } else {
      streamB += clean[i];
    }
  }

  // Case 1: Stream A is the full tracking number, Stream B is a duplicate prefix
  if (streamA.startsWith(streamB) && streamA.length >= 10) {
    return streamA;
  }

  // Case 2: Stream B is the full tracking number, Stream A is a duplicate prefix
  if (streamB.startsWith(streamA) && streamB.length >= 10) {
    return streamB;
  }

  // Safety Boundary: If there is any mathematical ambiguity, do not guess.
  return clean;
}



Edge Cases Handled:




Consecutive identical repeats: Standard duplicate bursts (AB1234AB1234) are collapsed separately using regex-based cycle detection.


Ambiguity Guard: If the decomposed streams do not cleanly form a prefix/subset relationship of each other, the algorithm leaves the raw string intact. We never guess and corrupt valid tracking numbers.






4. Architectural Lessons from Warehouse Software



Solving this bug highlighted three core engineering principles for industrial software:



1. Hardware Input is Dirty Data



Treat barcode scanners, serial scale interfaces, and optical sensors like unvalidated public API endpoints. Never write incoming hardware streams directly to persistent storage without validation and normalization layers.



2. Edge Processing over Cloud Latency



On a busy warehouse sorting line, workers process parcels at an intake rate of 1 parcel every 3 to 5 seconds. Sending video streams or raw scans to cloud endpoints for parsing introduces round-trip network lag. Real-time de-interleaving, dimensioning, and OCR must run locally on the client or edge device.



3. Progressive Isolation



Keep low-level hardware sanitization isolated from domain logic. The barcode normalization module lives as a pure, dependency-free TypeScript library covered by comprehensive automated tests, making it easily shareable across native Android wrappers and desktop web clients.





Wrapping Up



Building software for physical operations bridges the gap between clean code and dirty industrial realities. When optical imagers, thermal printers, and network dead-zones collide, robust client-side software architecture is what keeps the supply chain moving.




Have you encountered strange hardware quirks or keyboard-wedge edge cases in production? Share your war stories in the comments below!


![ ](https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/8ynp4afd7g7qazxknint.jpg)






We build TikTrake, an edge-first consolidation warehouse and cross-border freight management platform.