Top Tech Voices Podcast S2 E6: Smart Cities & Urban Tech

Carlo Ratti’s discussion on open-source architecture and treating environments as responsive, living organisms is a masterclass for hardware developers. His point that 'you can only change what you know' is the ultimate argument for deploying localized edge-sensing.

​I plan to use the Arduino Uno R4 to build an open-source, edge-computing nodes array for micro-climate and localized air quality tracking—directly mirroring the philosophy behind his 'City Scanner' project. By leveraging the Uno's processing power to handle real-time sensor data loops without relying on heavy external libraries, we can prototype the exact low-cost, decentralized cyber-physical systems needed to make our communities more adaptive. Exceptional interview, Farnell team!

Dipeshkachhi said:

I plan to use the Arduino Uno R4 to build an open-source, edge-computing nodes array for micro-climate and localized air quality tracking—directly mirroring the philosophy behind his 'City Scanner' project.

I recall that there was a study by University of Surrey involving the use of sensor box devices fitted to prams in order to sample the air quality as they were pushed around urban areas.

https://www.surrey.ac.uk/news/when-are-babies-prams-most-exposed-air-pollution

https://www.sciencedirect.com/science/article/pii/S0160412019345593#s0010

I think Bristol may have done something similar as part of their Smart Cities project. It was featured in a documentary programme but can't recall which one. I recall members of the public were invited to fit them to prams to build up a wider picture of daily  pollution exposure in urban areas.

 Those pram-mounted sensor studies from Surrey and Bristol are the absolute perfect real-world example of what Carlo Ratti means by a 'living, breathing data canvas.' Mobile citizen-sensing completely changes the game because it maps air quality at the exact breathing height of a human, rather than using a static municipal sensor mounted 10 meters up on a street lamp.

​The major engineering bottleneck with those early citizen-science deployments was data ingestion and power—they often relied on heavy, power-hungry logging shields or continuous cellular uploads that drained batteries rapidly.

​That’s exactly why I’m looking at the Arduino Uno R4 for prototyping this open-source node array. With its 32-bit RA4M1 Renesas processor, we have enough computational headroom to run localized data-filtering algorithms (like running average filters or threshold analytics) right on the edge. Instead of broadcasting massive streams of raw, noisy sensor data continuously, the node only needs to log or transmit compressed, cryptographically signed data packets when a significant pollution spike occurs.

​If we can keep the hardware low-cost and power-efficient enough to run off a small battery or energy-harvesting setup, deploying these onto public transit or citizen vehicles becomes completely viable. Have you ever experimented with localized data filtering on the Uno R4 to optimize battery life in remote sensor applications?

 Those pram-mounted sensor studies from Surrey and Bristol are the absolute perfect real-world example of what Carlo Ratti means by a 'living, breathing data canvas.' Mobile citizen-sensing completely changes the game because it maps air quality at the exact breathing height of a human, rather than using a static municipal sensor mounted 10 meters up on a street lamp.

​The major engineering bottleneck with those early citizen-science deployments was data ingestion and power—they often relied on heavy, power-hungry logging shields or continuous cellular uploads that drained batteries rapidly.

​That’s exactly why I’m looking at the Arduino Uno R4 for prototyping this open-source node array. With its 32-bit RA4M1 Renesas processor, we have enough computational headroom to run localized data-filtering algorithms (like running average filters or threshold analytics) right on the edge. Instead of broadcasting massive streams of raw, noisy sensor data continuously, the node only needs to log or transmit compressed, cryptographically signed data packets when a significant pollution spike occurs.

​If we can keep the hardware low-cost and power-efficient enough to run off a small battery or energy-harvesting setup, deploying these onto public transit or citizen vehicles becomes completely viable. Have you ever experimented with localized data filtering on the Uno R4 to optimize battery life in remote sensor applications?

You’ve highlighted the exact engineering trade-offs that keep sensor network designers up at night. The tension between data relevance (breathing height) and data integrity (controlled municipal calibration) is a classic cyber-physical systems dilemma.

​To your point about the "pram scenario"—where a jacket accidentally smothers a citizen's mobile node—you lose control over the physical operating environment. One way to mitigate this at the hardware/firmware level is implementing local multi-sensor cross-verification. For example, pairing the particulate matter (PM) sensor with a localized temperature, humidity, or ambient light sensor on the same node. A sudden drop in ambient light combined with an isolated micro-climate humidity spike inside a bag could flag that specific data window as "obstructed/unreliable" before it ever gets uploaded.

​Your critique of threshold-based logging is also incredibly sharp. A static "significant spike" threshold is a major blind spot for historical baseline analysis, especially if atmospheric baselines shift over time. A more robust alternative for the Uno R4 firmware would be a delta-based logging algorithm (Delta-Sigma modulation or slope-detection). Instead of waiting to cross an arbitrary hard ceiling, the node logs data whenever the relative rate of change (\frac{\Delta PM}{\Delta t}) exceeds a specific gradient, or when the value deviates from a moving local average. This preserves the subtle shifts needed for long-term historical modeling while still compressing data packets significantly.

​The tire-wear particulate matter (TWPM) issue from EVs is a fascinating and urgent variable. Because tire particles are typically larger and heavier (PM_{10} and microplastics) than highly airborne diesel soot (PM_{2.5}), their settling behavior is completely different. This makes mapping them at human breathing zones vs. street-lamp height even more critical, as heavy tire dust stays closer to the asphalt before being kicked up by passing vehicles.

​I really like Doug’s approach with the solar and supercapacitor combination for his weather station. For high-drain Edge AI or frequent RF transmission cycles, supercapacitors are unmatched in handling rapid charge/discharge cycles without the degradation issues of LiPo batteries in extreme outdoor temperatures. I’ll definitely dig deeper into Doug's "Just Encase Solar Super Capacitor" design—it sounds like the ideal power topology for a truly maintenance-free, localized urban node array.

Dipeshkachhi said:

Mobile citizen-sensing completely changes the game because it maps air quality at the exact breathing height of a human, rather than using a static municipal sensor mounted 10 meters up on a street lamp.

However, need to be mindful of the reliability and accuracy of the data.

A municipal sensor 10m up can be sited, installed, calibrated and maintained under a controlled set of conditions. However most people are probably not actually breathing the air at a height of 10m so the recorded data may be very accurate and consistent but not be all that meaningful.

Whereas with a citizen deployed sensor, you may potentially end up with more meaningful data being sampled at human height, but then lose control over accuracy. For example, the sensor was placed on the pram as instructed but then a garment or bag was placed directly on top of it affecting its reading throughout the capture period. 

Dipeshkachhi said:

the node only needs to log or transmit compressed, cryptographically signed data packets when a significant pollution spike occurs.

Perhaps need to be mindful that the 'significant spike' may have an unknown threshold or a value that reduces over time. If you want to be able to work with historical data in the future, then you may need to transmit data when values change, rather than when they cross a threshold.

I was listening to a podcast recently about rubber pollution from tyres which is now becoming an increasing concern. While filters have been cleaning up particulate matter from diesel emissions, apparently this is now being offset by increased particulate matter produced from tyre wear due to the increased weight and acceleration from EVs. It sounds like it is still relatively unknown whether rubber particulate matter is any more or less harmful than diesel combustion particulate matter.

Dipeshkachhi said:

Have you ever experimented with localized data filtering on the Uno R4 to optimize battery life in remote sensor applications?

I've not done much with the R4 as yet. Too busy watching/listening to podcasts...

Not R4 but Doug's weather station project used solar combined with supercapacitors for energy harvesting for sensor applications:

Just Encase Solar Super Capacitor