Water Management
Muslima Alia
Our journey towards building a highly accurate IoT water level monitoring system began with a simple question: how can we make users trust the readings they see?
We know that even minor inaccuracies could lead to poor decisions — overflow, dry runs, or wasted water. So, we didn’t settle for good enough. We aimed for over 99% accuracy. Achieving that meant rethinking every layer—starting from the sensors and signal noise, down to how we interpret data in real time. Our device, Waltr A, is also certified through Field Comparison Reports (FCRs), where its reading matched the actual water level with over 99% accuracy.
We understand that even small inaccuracies can lead to major issues—like tank overflows, dry runs, or wasted water. That’s why we didn’t settle for “good enough.” We aimed for over 99% accuracy. Achieving this required rethinking every layer—from reducing sensor noise to optimizing how we interpret data in real time. Waltr A’s performance was validated through Field Comparison Reports (FCRs), conducted by our internal field team. These tests confirmed that Waltr A’s readings consistently matched actual water levels with over 99% accuracy.
We tested our system through thousands of pump cycles, in varying weather, across different tank types, and with real-life user behavior. With each version, we improved our proprietary measurement logic, built smarter firmware, and developed custom tools to validate every reading. It was through this iterative, ground-up process that we refined the system to be as reliable in real-world rooftops as it is in controlled environments.
In this blog, we explain our approach to hardware design, the evolution of our firmware, the challenges we face, and how software plays a key role in achieving this level of precision.
The Hardware Evolution
When we first began developing our water level monitoring system, we started with the standard off the shelf ultrasonic sensor. It was easily available, highly affordable, and great for prototyping. For small-scale indoor experiments, it worked well and gave us a quick way to test our logic and measurement flow. However, once we moved beyond the lab and into real-world installations, the limitation became obvious:
Lack of Waterproofing: Even though the sensor doesn’t have to touch water to take readings, the environment inside a water tank is humid and prone to a lot of condensation. The exposed circuitry and open transducers meant it couldn’t handle humidity or water ingress. The sensor had to be placed directly on the tank lid, making it difficult to protect or reposition. Over time, these conditions led to corrosion, short circuits, and complete sensor failures in the field.
To overcome these challenges, we moved to a waterproof ultrasonic sensor. It meant a 10x increase in product costs but gave more reliability. Designed for outdoor use, it came with an IP66 coating that protected it from water and dust. Unlike the previous sensor, its sensor head could be installed separately from the control board, allowing for more flexibility in placement.
It also had a better range of up to 6 metres, making it suitable for deeper tanks. The sealed casing, waterproof cabling, and overall robust build made it far more durable in real-world environments.
This shift improved reliability and helped us scale confidently.
More Condensation, Foam and Spider webs:
As we continued doing more deployments, we came across diverse tank environments. Tanks where frequent spider web formation was happening, STP tanks where foam would form on top of the water layer, tanks where too much condensation was happening which created a lot of water vapour in the air.
Switching to Pressure Sensor:
Pressure sensors seemed an obvious choice, as they would not have any direct impact due to these environmental conditions. However, it had two major drawbacks:
The dust and debris forming inside the tank will clog the pressure sensing holes of the sensor, which will result in the sensor giving wrong readings. As a result, it will require frequent cleaning of the sensor holes, which will be an unpleasant experience for the customers.
There will be a significant increase in the product cost which may make it a non-starter for many of our customers.
Radar to the rescue:
While ultrasonic sensors served us well in more than 90% of the cases, we still wanted better results. As we aimed to achieve the level of accuracy of a pressure sensor, without its drawbacks or cost increase, we zeroed in on mm Wave Radar as our sensing technology.
“Getting this level of control over the hardware changed everything,” shares our hardware engineer, S Maheshwaran. “With ultrasonic, we could only use what was given. But with radar, we design and program the sensor ourselves—so we know exactly how it behaves and can tune it for any tank.”
This shift wasn’t just about changing the sensor—it was about taking full control of the solution. Off-the-shelf ultrasonic sensors give us little room to adapt or optimize. The radar allowed us to start from the ground up. Today, we build our radar sensors entirely in-house—from material selection and PCB design to antenna layout—giving us consistent, high-precision performance across all tank types and conditions.
Each radar sensor is paired with a low-power microcontroller that balances processing capability with energy efficiency. Power regulation is built into the board. The device wakes up every 30 seconds, activates only the necessary components to take the reading, then goes back to sleep—conserving battery life without compromising accuracy.
We’ve tested the radar system in rooftop tanks, underground sumps, and high-rise reserves—across different shapes, materials, and mounting conditions. Unlike ultrasonic sensors, which were rigid and failed in non-standard tank environments, radar gave us the flexibility to adapt. With full control over the firmware and hardware, we could optimize the sensor for any environment, indoors or outdoors, and in extreme weather conditions.
One of the best things about radar is that it can be adjusted to work perfectly with different type of water tank for example, if a tank is tall, short, round, or square the radar can be set up to match it exactly. Ultrasonic only tells us the distance to the water, but radar gives more accurate data because it can be fine-tuned.
Our device,Waltr A is also certified through Field Comparison Reports (FCRs), where its reading matched actual water level with over 99% accuracy. The result is a compact, rugged, and intelligent hardware unit—engineered not just to survive challenging conditions, but to perform reliably and accurately over time. Every layer, from antenna to microcontroller, was designed to work in harmony—with energy efficiency, precision, and reliability at its core.
One of radar’s biggest advantages is its ability to be precisely calibrated. Unlike ultrasonic sensors that return only distance data, our radar system allows for detailed adjustments based on the tank size, shape, and height. This flexibility has been crucial in achieving over 99% accuracy—as validated by our Field Comparison Reports (FCRs), which match actual measured volumes with sensor data.
The Software Transformation
Smarter Sleep and Power Efficiency
We implemented adaptive sleep cycles that let the device wake up every 30 seconds, take a reading, and connect to Wi-Fi only when it detects significant water level changes. Since Wi-Fi transmission draws power for more than 10 seconds, avoiding unnecessary communication leads to significant energy savings—especially in battery-powered setups. This is particularly useful during the rainy season, when solar charging is less and energy efficiency becomes critical. We also enabled backend control of sleep timings, allowing us to remotely adjust device behavior based on the environment or usage patterns, without requiring a firmware update.
Ensuring Data Reliability
To ensure data reliability, we integrated statistical filtering into the measurement process. Every time the device takes readings, it takes multiple samples and sends the most likely reading—not a raw point.
This approach eliminates outliers caused by environmental noise or sensor glitches and results in cleaner, more stable data. If the device fails to get a valid reading, it doesn’t give up immediately. It attempts the measurement up to three times, each time collecting nine samples. Even if only a few of those samples are accurate, the filtering helps extract a reliable value. This makes our system robust even in challenging signal conditions.
Remote Configuration and OTA Updates
We introduced full remote configurability, including control over sleep durations, transmission thresholds, and Over-The-Air updates. OTA checks were optimized to balance battery conservation and system reliability.
Communication Resilience and Data Backup
We also built fallback mechanisms. If data transmission is missed, devices will retry or await manual triggers. When a device cannot connect to Wi-Fi or the server, it does not discard the data—it stores readings locally and uploads them once connectivity is restored.
In future, we can configure whether to upload data one-by-one or in batches of 10 or 20 readings, buffering up to 40 minutes of data if needed. We can also manually trigger uploads or reconfigure sleep cycles in real time.
System health is also continuously monitored. Inactive or misbehaving units are flagged, and alerts are triggered for unusual patterns such as repeated incorrect readings—allowing us to detect and resolve issues before they affect the user.
Architectural Improvements for Scale
Earlier, each device sent over 600 readings everyday to a central server, which batched and processed data hourly. While functional, this introduced latency in detecting errors, any problem in the chain could affect the entire system. We restructured our architecture to assign each device its own isolated processor. This isolation meant failures in one device didn’t impact others, and troubleshooting became faster.
Additionally, we adopted staggered upload timings to spread device activity across the hour, reducing server load, lowering costs, and improving scalability and performance.
Smarter Data Filtering
Processing the data came with its own set of challenges. Inflow during pump operation creates sharp increase, while outflow from user consumption tends to be slower and smoother. Initially, we used z-score filtering to clean up the data, but it wasn’t suitable—water usage patterns aren’t normally distributed. We moved to rolling medians and then to range based filtering, which doesn't assume a specific distribution and proved more effective.
We created thresholds to account for the unique behavior of tanks in the field. For example, if a value looked like an outlier but occurred repeatedly during a pump cycle, it was likely real and should be preserved. This allowed us to eliminate noise without discarding any meaningful data.
You might wonder—why 99% and not 100%? In real-world conditions, perfect accuracy is nearly impossible. Factors like environmental noise, installation differences, or signal interference can introduce tiny errors. But we’ve designed our system to minimize those as much as possible. Reaching and maintaining over 99% accuracy wasn’t just a technical goal—it was essential for trust.
This journey has been shaped by continuous improvement, technical challenges, and the drive to build an IoT system that works reliably in real-world conditions. From early prototypes with ultrasonic sensors and Arduino boards to a robust platform powered by radar sensors and ESP-IDF firmware, every stage brought new lessons. Through rigorous testing, real-time data analysis, and thoughtful hardware decisions, we’ve developed a smarter, more scalable product. These efforts not only strengthened our current IoT water monitoring system but also laid the groundwork for future advancements in sensing, control, and automation.