Definition
PSM (Power Saving Mode) is a deep sleep mode used in technologies such as NB-IoT. During idle periods, the terminal enters PSM, turns off the RF transceiver, and only keeps the RTC wake-up function active, resulting in extremely low power consumption, allowing the battery to last for several years. It is suitable for low-frequency reporting scenarios such as water meters and manhole covers.
Working Principle
After the terminal has been idle for a certain period of time (defined by T3324), it enters PSM, shuts down the RF, and only the RTC runs, making it unreachable by the network. Wake-up is initiated by the terminal itself (such as for scheduled reporting) or triggered when T3412 (Active Timer) expires, exiting PSM and entering eDRX to listen for paging. The duration of PSM can be configured, up to several weeks. Before entering PSM, the terminal informs the network, and the network caches downlink data to be delivered once the terminal wakes up.
Used in conjunction with eDRX
eDRX stands for Extended Discontinuous Reception, where the terminal periodically and briefly wakes up to listen for paging; PSM stands for Power Saving Mode, which is deep sleep and cannot be paged. Pure uplink scenarios (such as meters reporting daily) can use only PSM; scenarios requiring downlink wake-up need to combine eDRX. The PSM+eDRX combination can balance battery life and downlink latency, NB-IoT typically can support a battery life of 5–10 years.
Application Recommendations
Smart meters, environmental sensors, manhole monitoring, and other low-frequency reporting devices are suitable for enabling PSM. Scenarios requiring real-time downlink should shorten PSM or disable it. The platform side needs to tolerate the downlink latency caused by PSM or use alternatives like SMS. PSM parameters are configured via NAS signaling and need to match network operator support.
Sleep Timers and the Paging Gap
Once the radio is off, the network cannot page
The technical meaning of this power-saving mode is that the device tells the network, of its own accord, not to look for it for a while. After the idle timer expires, the radio, the baseband, and most of the clocks turn off. Only a low-power clock that can wake the device remains. During that time the device hears nothing on the paging channel. Downlink data can only sit on the network side and be delivered when the device returns to connected state on its own. What is saved is listening current. The price is that the device cannot be reached on the downlink. Average current can fall to the microamp range because the duty cycle of transmitting and listening is pushed very low, not because the chip becomes more economical while it is working.
Two timers set the rhythm. The shorter one is the tail during which the device is still willing to listen for paging before deep sleep. The longer one roughly decides how soon the device must come back and report to the network. Too short a tail, and the network does not have time to send downlink that has just arrived. Too long a period, and the buffer may be discarded. The device wakes and sees only "no data." Both values have to be accepted by the network. Writing them into the module does not mean the cell has agreed. If they are rejected, the device runs on the values the network assigned, and the battery budget no longer holds.
Suited only to a device that wakes itself to report
It suits meters and slow sensors whose reading is uploaded when the device decides, and that can accept downlink hours later. Commands that must arrive immediately — open a door, stop a machine, change a threshold remotely — cannot rest on the assumption of deep sleep. If the service occasionally needs faster downlink, shorten the sleep or switch to a mechanism that still listens for paging on a cycle, and design the application for the worst-case wait. Do not design it for the delay while the device is awake. Unlike a private LPWAN, the sleep contract here is signed with the carrier core network. Changing cell, changing the SIM, or a parameter refresh on the network side can rewrite the timers.
Before use, confirm that the module firmware, the SIM, and the current cell all support it, and check the negotiated result in the log. Do not look only at the configuration file. Report messages should carry their own timestamp, because the time they reach the platform is later than the sample time by one sleep period. If the platform draws the curve from arrival time, the shape of the curve changes whenever the sleep length changes.
T3324 and T3412 set the wrong way around
The most common error is to swap the two timers, or to set both to extremes. Someone who wants to save power stretches the wake period a long way and at the same time asks the platform to open and close a valve at any moment. The command waits in the network until it expires. Someone who wants control turns sleep off. The battery is consumed as if the device were listening continuously, and the site is undervoltage within a few months. Another error is a dense application-level heartbeat wrapped on top. Sleep is broken by the device's own heartbeat as soon as it is established. The radio almost never turns off. The configuration looks as if power saving is on, and the current waveform is full.
Retries after a failed attach also have to be rate-limited. A manhole cover in poor signal that reattaches immediately on every failure spends the benefit of sleep on retry current, and it also loads the cell. Back off longer after a failure, and record the last successfully negotiated timers in the diagnostics of the communication module, so "the network did not grant power saving" can be told from "the application never let it sleep." Finally, do not estimate underground battery life from current measured at full signal in the lab. In weak signal, transmit time grows. The same sleep period then yields a shorter battery life.