Data Loggers

What is a Data Logger

A data logger is an instrument that automatically measures and stores sensor data over time, without requiring a connected computer to keep running. Functionally it sits between a simple recorder and a full data acquisition system: it adds intelligence (a microprocessor, a clock, programmable scan logic, and onboard storage) to the raw act of measurement, so that the recording continues autonomously once configured. A logger turns a physical quantity such as temperature, strain, voltage, or current into a time-stamped digital record that an engineer can later download and analyze.

Every data logger contains four functional blocks. First, the signal-conditioning front end scales, filters, and protects incoming sensor signals, providing excitation for bridges and RTDs, cold-junction reference for thermocouples, and shunt resistors for current loops. Second, the analog-to-digital converter (ADC) turns the conditioned analog voltage into a digital number, and its resolution and noise floor set the achievable accuracy. Third, the processor and real-time clock run the measurement schedule, apply linearization and engineering-unit conversion, and time-stamp each record. Fourth, non-volatile memory (flash, SRAM with backup, or a removable SD card) retains the data through power loss, while a communication interface exports it.

The practical boundary that separates a logger from a data acquisition system is autonomy. A DAQ card streams samples to a host PC at high rates and depends entirely on that host for storage, processing, and display, so it stops the moment the computer stops. A logger buffers and stores its own data, and many modern instruments straddle the line: the Pico ADC-24, for example, is PC-connected yet buffers locally, while the Campbell Scientific CR1000Xe scans at up to 1000 Hz, stores to onboard flash and microSD, and serves data over Ethernet and Modbus. The deciding question is what happens when the network or the operator disappears.

Data loggers became practical when low-power microcontrollers and solid-state memory matured in the 1980s, replacing the strip-chart and circular paper recorders that had dominated process recording for decades. The shift from paper to digital storage brought time-stamping precision, remote download, alarm logic, and the audit-trail capability now mandated for regulated industries. Today the category ranges from sub-dollar single-use temperature tags shipped inside vaccine boxes to ruggedized field stations that survive arctic and desert deployment for years on a small solar panel.

Four engineering metrics determine logger quality across that entire range: measurement accuracy (including the dominant cold-junction or excitation error for the sensor in use), resolution (set by the ADC and its noise floor), channel architecture (single-ended versus differential, isolation, and expansion), and autonomy (memory depth, battery life, and ingress protection). These four jointly determine whether a logger will actually deliver trustworthy data over the intended deployment, and they form the backbone of the chapters that follow.

Data Logger Types

Data loggers are most usefully classified by how they connect, store, and are accessed, because that architecture dictates where they fit in a measurement workflow. Five families cover the great majority of industrial and scientific deployments: standalone, PC-based, wireless/IoT, web-configurable, and modular high-channel-count systems. The table below summarizes the core trade-offs, and the paragraphs that follow describe each family in depth.

Standalone loggers are the simplest and most numerous. They record directly to internal memory or a removable SD card and are physically retrieved or briefly connected to offload data. A compact example is the Onset HOBO MX2301A, an IP67-rated weatherproof temperature and humidity logger with Bluetooth offload, rated plus-or-minus 0.2 degrees Celsius from 0 to 70 degrees C. Standalone units win on cost, battery life, and ruggedness, and they dominate cold-chain, building-science, and field-survey work where the priority is unattended endurance rather than live data.

PC-based loggers connect to a host computer over USB or Ethernet and rely on it for storage, configuration, and display, while still buffering locally to bridge brief interruptions. They excel where the application needs high resolution and continuous visualization at a bench: the Pico Technology ADC-24 is representative, offering up to 24-bit resolution, galvanic isolation from the PC, and up to 16 single-ended or 8 true differential inputs across seven ranges from plus-or-minus 39 mV to plus-or-minus 2500 mV. The category trades autonomy for software-rich analysis and low per-channel cost.

Wireless and IoT loggers add a radio (Bluetooth, Wi-Fi, LoRaWAN, or cellular) so that data reaches a phone, gateway, or cloud platform without a physical visit. They suit distributed and mobile assets such as refrigerated trucks, warehouse zones, and remote infrastructure, where the cost of manual download would be prohibitive. Local buffering remains essential so that a radio outage does not create a data gap, and battery budget becomes the dominant design constraint because transmission is far more power-hungry than measurement.

Web-configurable loggers embed a web server so that setup, viewing, and download happen entirely through a browser with no installed client software, easing cross-platform and remote operation. The Yokogawa modular GM10 is a clear example: a scalable, web-configurable acquisition station with per-channel A/D conversion, expandable up to 100 channels, designed for industrial and laboratory plant DAQ. Modular and rack systems extend this to the largest deployments. The Graphtec GL840 provides 20 isolated analog channels expandable to 200, and Campbell Scientific and Yokogawa platforms scale similarly for structural testing, power generation, and large pilot plants where hundreds of isolated channels and high inter-channel withstand voltage are mandatory.

ADC and Channel Architecture

Two architectural choices dominate logger accuracy and capacity: the analog-to-digital converter that digitizes the signal, and the input topology that connects sensors to it. Resolution is set by the number of ADC bits and, critically, by the noise floor that reduces the usable count below the nominal figure. Channel capacity is set by whether inputs are single-ended or differential and by how the front end isolates one channel from another. The table below contrasts the key engineering metrics of the two mainstream ADC families used in loggers.

Resolution versus accuracy is the most misunderstood pair on a logger datasheet. Resolution is how finely the ADC subdivides the input span: a 16-bit converter creates 65,536 steps, so a plus-or-minus 10 V range resolves about 0.3 mV per step, while a 24-bit converter creates over 16 million steps. The Graphtec GL840 uses a 16-bit sigma-delta ADC with an effective resolution near 1/40,000 of full scale, and the Pico ADC-24 and Campbell CR1000Xe reach 24-bit. Accuracy, by contrast, is how close the reading is to truth, and it is bounded by gain error, offset, linearity, temperature drift, and noise. A logger can have abundant resolution and still be inaccurate; the useful, noise-free resolution is always several bits below the nominal count, which is why specifying by bit count alone is a beginner error.

Sigma-delta converters oversample at a high rate and digitally filter the result, trading speed for resolution and excellent rejection of mains-frequency interference. This makes them the default for temperature, strain, and low-level millivolt logging, where channels change slowly and noise rejection matters more than bandwidth. Their effective sample rate per channel is typically up to roughly 1 kHz, which is ample for thermal and structural monitoring but inadequate for transient capture. Successive approximation (SAR) converters digitize one sample per clock cycle without oversampling, reaching far higher rates at moderate resolution, which suits multi-channel scanning of fast signals such as vibration and pressure transients.

Single-ended versus differential inputs determine both noise immunity and channel count. A single-ended input measures one signal wire against a shared ground and offers the most channels per terminal block, but it is exposed to ground-loop currents and common-mode noise. A differential input measures the voltage between two dedicated wires and rejects common-mode interference, at the cost of consuming two terminals per channel. The hardware reflects this directly: the Campbell CR1000Xe provides 16 single-ended or 8 differential channels from the same terminals, and the Pico ADC-24 provides up to 16 single-ended or up to 8 true differential inputs. Use differential for thermocouples, bridges, and long cable runs; reserve single-ended for short, clean, grounded signals where doubling capacity is worth the noise risk.

Channel-to-channel isolation is the third architectural axis, decisive when channels sit at different ground potentials or measure off-ground sources such as shunts and powered or grounded thermocouples. The Graphtec GL840 isolates every analog channel from the others and from power ground, with maximum input ratings of 60 V peak-to-peak between terminals and up to 600 V peak-to-peak between channels on the withstand-voltage model. Isolation prevents a fault on one channel from corrupting or destroying the rest, and it is mandatory in power, automotive, and high-energy test environments. Non-isolated multiplexed front ends are cheaper and faster but assume all signals share a common reference.

Sensor Inputs and Standards

A logger is only as accurate as the way it conditions each sensor type, and each input class brings its own standard, error sources, and wiring rules. The major input types in industrial and scientific logging are direct voltage, current loop, thermocouple, RTD, bridge (strain and load), and pulse or frequency. The table below maps each input to its governing standard and the error term that usually dominates, before the discussion that follows.

Thermocouples are the most common logger input and the most subtly error-prone. A thermocouple generates a voltage proportional only to the temperature difference between its measuring junction and the terminal block, so the logger must measure that terminal temperature with a built-in reference sensor and add it back: this is cold-junction compensation. Any error in the cold-junction sensing transfers directly, a 0.5 degrees Celsius reference error adds 0.5 degrees C to every channel. IEC 60584-1 defines the tolerance of the wire itself, with Class 1 the tightest; a Type K Class 1 element is typically within roughly plus-or-minus 1.5 degrees C or 0.4 percent over much of its range. Loggers such as the Graphtec GL840 support Types K, J, E, T, R, S, B, N, and W, each with its own range and sensitivity.

RTDs, principally the platinum PT100, offer better stability and linearity than thermocouples and are governed by IEC 60751. Class A tolerance is plus-or-minus 0.15 degrees Celsius at 0 degrees C, widening to plus-or-minus 0.35 degrees C at 100 degrees C per the formula plus-or-minus (0.15 + 0.002 times the temperature magnitude in degrees Celsius). Class B is plus-or-minus 0.3 degrees C at 0 degrees C, the price-performance default for industry. The dominant logger-side error is lead resistance, which is why precision RTD logging uses 3-wire or 4-wire connection to cancel cable resistance; Class A accuracy is not valid in a 2-wire configuration because the lead resistance uncertainty swamps the tolerance.

Current-loop and voltage inputs are the workhorses of process logging. A 4-20 mA loop is read by passing it through a precision shunt resistor and measuring the resulting voltage, so the shunt tolerance dominates accuracy; the Campbell CR1000Xe supports native 4-20 mA and 0-20 mA current inputs alongside its plus-or-minus 5 V analog range. Direct voltage inputs span from low-level millivolt thermocouple-class signals to high-voltage ranges, the Graphtec GL840 covering 20 mV to 100 V full scale across selectable ranges. Bridge and pulse inputs round out the set: strain gauges and load cells need stable ratiometric excitation (load cells follow OIML R60 accuracy classes), while pulse and frequency channels need correct threshold and debounce settings to count tipping-bucket rain gauges, flow meters, and energy pulses without miscounts.

Compliance standards govern whether logged data is legally usable. EN 12830:2018 specifies requirements and test procedures for temperature recorders used in transport, storage, and distribution of temperature-sensitive goods between -80 and +85 degrees Celsius, and its higher accuracy class demands plus-or-minus 0.2 degrees C over the -25 to +7 degrees C band where the 1999 edition allowed plus-or-minus 0.5 degrees C. In regulated pharma and food, FDA 21 CFR Part 11 requires audit trails, access control, and tamper-evident electronic records, met by 21 CFR-compliant logger families from Lascar and others. Underpinning both, ISO/IEC 17025:2017 accredited or NIST-traceable calibration provides the documented, unbroken chain back to a national standard that keeps every reading defensible.

Key Specification Parameters

Logger datasheets list dozens of lines, but only a handful drive a sound selection decision: number and type of channels, ADC resolution, measurement accuracy, scan or logging rate, input ranges, memory depth, power and battery life, operating temperature, and ingress protection. Each is decoded below, with verified figures from representative instruments so the magnitudes are concrete rather than abstract.

Channels and ADC resolution set capacity and granularity. State the count in both single-ended and differential terms, because they differ: the Campbell CR1000Xe offers 16 single-ended or 8 differential, while the Graphtec GL840 offers 20 isolated channels expandable to 200. Resolution ranges from the 16-bit sigma-delta of the GL840 (effective near 1/40,000 of full scale) to the 24-bit converters in the Pico ADC-24 and CR1000Xe. Always weigh resolution against the noise floor, since the usable, noise-free count is several bits below nominal.

Measurement accuracy is the parameter buyers most often misread, because it depends on input type. The CR1000Xe specifies analog voltage accuracy of plus-or-minus (0.04 percent of measurement plus offset) at 0 to 40 degrees Celsius, widening to plus-or-minus (0.06 percent) at -40 to +70 degrees C and plus-or-minus (0.08 percent) at the extended -55 to +85 degrees C range. For thermocouple inputs the total error is dominated by cold-junction compensation and wire tolerance, not the ADC, so a Type K reading near plus-or-minus 0.5 to 1 degrees C reflects IEC 60584 wire plus reference error. Read accuracy as a system specification across the operating temperature band, never as a single bench number.

Scan and logging rate separate fast and slow loggers. The CR1000Xe scans at up to 1000 Hz, suitable for some transient work, while sigma-delta multi-channel loggers like the GL840 sample as fast as 10 ms per interval for voltage-only channels and slow down for thermocouple measurement. The logging interval, distinct from the instantaneous scan rate, is what an engineer programs for a deployment and drives both memory consumption and battery life.

Input ranges must bracket the signal with headroom. Precision low-level loggers offer narrow ranges for resolution: the Pico ADC-24 provides seven ranges from plus-or-minus 39 mV to plus-or-minus 2500 mV. Multi-purpose loggers span wide: the Graphtec GL840 covers 20, 50, 100, 200, 500 mV and 1, 2, 5, 10, 20, 50, 100 V full scale. As with pressure and other analog instruments, keep the working signal within the lower portion of the range only when resolution allows, and never let a transient exceed the input limit.

Memory, power, temperature, and ingress govern autonomy. The CR1000Xe carries 4 MB SRAM plus 72 MB flash, expandable to 16 GB via removable microSD, runs on 10 to 36 Vdc, and is rated to the extended -55 to +85 degrees Celsius band. Field loggers add ingress protection: the Onset HOBO MX2301A is IP67-rated for outdoor exposure. The relationships that matter are simple: memory consumed equals channels times bytes per record times samples per day times days, while battery life falls as logging interval shortens and radio use rises. For most thermal deployments, battery and ingress, not memory, set the practical limit.

Selection Decision Factors

To turn the preceding five chapters into a specific model, work through the decision sequence below in order. Most selection mistakes come not from a single wrong answer but from deciding a downstream parameter before the upstream constraint that bounds it. These nine steps double as a fixed RFQ template.

1. Measurand and sensor type: Decide first what you are measuring (temperature, strain, voltage, current, pulse) and the sensor that produces it (thermocouple, RTD, bridge, transmitter), because this fixes the input class, the governing standard, and the dominant error term before anything else.
2. Channel count and topology: Count present and future channels, then decide single-ended versus differential per Chapter 3. Specify required channel-to-channel isolation and withstand voltage when channels sit at different potentials, and verify the expansion path if the project may grow.
3. Accuracy and resolution target: State the smallest change you must detect and the accuracy you must guarantee across the full operating temperature band. Match ADC resolution and converter type (sigma-delta for low-level and slow, SAR for fast multi-channel) to that target, weighed against noise floor, not nominal bits.
4. Sampling and logging interval: Set the scan rate to resolve the fastest process change, and the logging interval to the slowest rate that still captures it, balancing memory and battery. Enable alarm-triggered fast logging only where transient detail is genuinely needed.
5. Memory and autonomy: Compute memory as channels times record size times samples per deployment, then size storage with margin. Confirm battery life or power source over the full unattended period, since autonomy, not capacity, usually limits field loggers.
6. Connectivity and access: Choose standalone download, PC-based streaming, wireless or cellular telemetry, or web-configurable browser access per Chapter 2. For plant integration, confirm Modbus, Ethernet, SDI-12, or fieldbus support and protocol compatibility with the host system.
7. Environment and ingress: Match operating temperature (the CR1000Xe extended band reaches -55 to +85 degrees Celsius), humidity, vibration, and ingress protection (IP65, IP67, or higher for outdoor and washdown) to the deployment, since a logger that fails environmentally produces no data at all.
8. Compliance and calibration: Identify required standards: EN 12830:2018 for cold chain, FDA 21 CFR Part 11 for regulated electronic records, OIML for legal metrology, plus ISO/IEC 17025 accredited or NIST-traceable calibration certificates and a defined recalibration interval.
9. Total cost of ownership (TCO): Sum purchase price, software licensing, annual calibration, battery and consumable replacement, and the cost of any data gap. A cheaper logger that drifts out of tolerance or loses data during a transport leg can cost far more than its price difference in a single failed shipment.

One dimension is routinely underweighted at purchase: manufacturer serviceability, meaning calibration-lab access, firmware update support, spare-sensor availability, and software longevity over a 5 to 10 year service life. Campbell Scientific, Yokogawa, Graphtec, Pico Technology, Onset, and Lascar each maintain established support and calibration ecosystems, which is why they remain reliable choices for projects where the logger must keep delivering defensible data long after the original purchase order is closed.