Find the gaps: Boost your network’s energy performance with time domain techniques

5G New Radio (NR) technology is designed to be ultra-lean compared to previous cellular standards including 4G (LTE). It boosts energy efficiency by minimizing the periodic transmission of the always-on reference signals that help the UE to secure cell coverage and a good connection with users. However, NR still consumes considerable energy per transmission due to the increasing number of antennas, larger bandwidths, enhanced processing, and higher frequencies. In an earlier blog post, Network energy consumption modeling and energy saving technologies, we shed light on energy-saving techniques in the spatial, time, power, and frequency domains and presented our evaluations of them based on an NR base station energy model. 

Here, using the same model, we focus on time-domain techniques applicable to devices in both connected and idle/inactive modes. We will show that establishing sufficiently long transmission and reception gaps will result in significant energy savings (up to 75 percent) in an NR base station. 

Small things do matter

In one of his Sherlock Holmes stories, Sir Arthur Conan Doyle stated, "It has long been an axiom of mine that the little things are infinitely the most important." NR technology is a good example of this principle. NR base stations (gNBs) transmit a relatively small-bandwidth four-symbol synchronization signal block (SSB) with a typical periodicity of 20ms. This is much less frequent compared to previous systems like LTE, which had always-on wide-band reference signal transmissions. 
Mobile devices (referred to as user equipment, or UE) rely on this transmission scheme for various critical functions, such as searching for cells, measuring signal strength, connecting to the network, and maintaining the connection. Moreover, a gNB regularly monitors the physical random access channel (PRACH) for potential transmissions from UEs, typically every 10 to 20 ms. This is done to detect any UEs attempting to access the system through the gNB. When incoming calls are made to UEs, other gNBs within their coverage areas occasionally transmit paging messages to reach UEs that are located in that area. The paging occasions (POs) for reaching various UEs are by design spread in time and may therefore happen at many possible points in time. 

At first glance, transmitting occasional paging messages in some POs, a few SSB symbols once every 20ms, and listening for PRACH may seem insignificant. However, even small transmissions/receptions can deprive the gNB of longer sleep opportunities, resulting in increased energy costs. 

Why are small transmissions/receptions costly?

Assuming the energy model described in our earlier blog post Network energy consumption modeling and energy saving technologies, one thing is clear: Putting a gNB into sleep mode during silent periods, when no downlink/uplink (DL/UL) transmissions/receptions are ongoing, is one of the most effective ways of conserving energy. Note that there are different levels of sleep modes; micro, light, and deep sleep as depicted in Figure 1. The deeper the sleep level, the more hardware blocks are turned off, resulting in lower energy consumption. However, each sleep mode is associated with a corresponding transition period, that is, the time it takes for the gNB to enter the associated sleep mode, wake up, and be prepared for active transmissions/receptions. 

Figure 1: Shows the different sleep modes of a gNB as specified by the NR advanced standards.

This means that for the base station to achieve a deeper level of sleep, the transmission/reception gaps need to be large enough to allow for the associated transition time. But these gaps don’t come naturally! The occasional SSBs, PRACH, and paging-related activities are distributed independently over time and collectively create enough interruptions to make predictable longer gaps unfeasible, thereby preventing the possibility of deeper sleep modes. Not to mention that each of these activities may individually be swept (repeated back-to-back) in time in multi-beam deployments. A service provider could create gaps by adjusting the frequency of transmissions/receptions. However, doing so could compromise the system’s responsiveness and impact overall performance. How do we tackle this issue?

Make them, but break them when needed

Imagine if you could configure generous transmission gaps as a starting point but override or compensate for them easily, whenever needed. This concept is at the heart of discussions in the 5G Advanced standards. It’s like what parents constantly tell their children: “Turn off the lights when no one is in the room,” or “Don’t let the water run while brushing your teeth.” The same principle applies here: avoid wasting resources when they are not needed, but ensure they are available when required. If no device requires SSBs, then don't transmit them. If no device is accessing a cell or a frequency band at the moment, then do not listen to the associated PRACH resources. But, as soon as there is a demand, ensure you provide the necessary resources. The critical questions are: where do you find the occasions for creating these gaps and how do you compensate for them swiftly without compromising performance? 

The answer is somewhat different depending on the scenario and resources involved, whether there are devices in connected mode or not, whether a potentially connected mode device is being served on a primary cell (PCell), or a secondary cell (SCell) in a carrier aggregation (CA) scenario. Let’s go through some of these scenarios below, separately for PCells and SCells.

PCells

First, let’s clarify what a PCell is and which of its functions we want to address. A PCell is the main cell that a UE uses for communication and typically used for control signaling and initial access. It is essential for establishing a connection between the UE and the network (for example, paging and random access) and is involved in various procedures such as handovers. It also serves as the reference cell in CA. In this context, we are targeting functions that interrupt PCell’s sleep even during low loads, specifically transmissions related to paging and SSB, and receptions related to PRACH. Let’s address these functions one by one.

PCell and SSB transmission: Do it when necessary, never out of habit

SSBs are currently designated as always-on transmissions on a PCell. When a UE conducts its initial cell search and actively seeks its first cell to connect to the network, SSBs on coverage cells must be provided every 20ms irrespective of cell traffic levels. Ensuring compliance with this timing is crucial in deployments to prevent UEs from prematurely abandoning searches, potentially leading to coverage holes.

However, this requirement does not dictate such dense SSB provision across all cells in a network. In deployments featuring overlapping cells (such as coverage and capacity cells exemplified in Figure 2), initial cell search functionality beyond the coverage cell is not essential. Once a UE has camped on the coverage cell, it can utilize neighboring cell information to learn about SSB transmission schemes, like less frequent transmissions (up to 160ms) in capacity cells. With this knowledge, UEs spend sufficient time searching for neighboring SSBs and can smoothly transition to capacity cells. This optimization allows for reduced SSB transmission rates in capacity cells as long as overlapping coverage cells maintain dense SSB provision.

Figure 2: Network deployment including two Pcells where a coverage cell is overlapping a capacity cell

It is important to note that adjusting the frequency of SSB transmissions — to make them less frequent can result in substantial energy savings in the network, allowing the gNB hardware to enter deeper sleep modes. Our evaluations, as exemplified in Figure 3, indicate that about 16 percent of energy can be saved by adjusting the periodicity of SSB, System Information block Type 1 (SIB1), and PRACH reception from 20ms to 40ms.

Figure 3: NW energy saving when SSB periodicity is increased from 20ms to 40ms.

So, why isn't this done today? The challenge lies in current 5G standards. Adjusting the periodicity of SSBs and PRACH can only be done semi-statically using higher-layer radio resource control (RRC) signaling. This process is slow, involving broadcast or dedicated transmissions that take effect with a delay, which does not allow swift adjustments. If we reduce a setting, like SSB transmission frequency, we need the ability to quickly compensate and provide SSBs when necessary. Otherwise, it could negatively impact system performance.

Therefore, as part of the advanced 5G standards (3GPP Rel-19), we propose a dynamic approach to SSB provision. For UEs in RRC-connected mode, SSB transmissions can be adjusted dynamically. Initially, SSBs can be provided sparsely (for example, every 160ms), serving as a baseline. Additional SSBs can be activated on-demand as needed, such as before/during handover or for positioning services like enhanced cell-ID (E-CID) and so on. Similarly, for UEs in RRC-idle mode, a longer baseline SSB periodicity (for example, greater than 20ms) can be set. Additional SSBs can be activated selectively based on demand. Formally, sparser SSB transmissions (up to 160ms) are feasible under current specifications, especially for capacity cells overlapped by a coverage cell. However, in practical deployments, cells with such low-rate SSB configurations are not typically deployed. This is because certain UEs in RRC idle mode, particularly those within poor coverage, may require multiple SSBs (up to three) in their loop convergence process to decode paging messages effectively. Figure 4 illustrates this scenario, where the UE would need to wake up more than 320ms before the PO to process the necessary SSBs, had they been transmitted with a 160ms periodicity. Such a configuration would significantly drain the UE battery, needing it to wake up so far ahead of the PO each time.

Figure 4:Timeline of RRC idle/inactive UE in need of multiple SSBs before the PO.

Instead, as illustrated in Figure 5, we expect further improvement from techniques where additional SSBs could be transmitted on demand before the POs for 5G advanced (3GPP Rel-19) UEs in RRC idle/inactive mode, alongside a baseline SSB transmission rate. This approach ensures backward compatibility, allowing legacy UEs to rely on the baseline SSB transmissions. It would then be feasible to deploy cells configured with low-rate SSB transmissions, addressing  practical concerns.

Figure 5: Timeline of RRC idle/inactive UE in need of multiple SSBs before the PO if SSB adaptations are done. The UE may choose to wake up closer to the PO when additional SSBs are provided.

Pcell and paging occasions: Omit unnecessary spread

Paging is another area currently under discussion for potential adaptations in 5G advanced. In the current specifications, POs of various UEs are uniformly distributed across the frames over time. This spreading of POs aims to balance the paging load in a cell from the gNB perspective while ensuring that paging messages are received by the UEs within the configured paging cycle.
From the network energy-saving perspective, however, this framework does not provide many opportunities for a gNB to sleep. In scenarios with high paging loads—especially when paging messages are escalated due to the UE’s lack of response in the initial paging attempts and broadcast across multiple cells within a UE registration area—gNBs rarely have opportunities to enter the deeper sleep modes. 

Greater savings require longer periods of consecutive sleep. A potential solution is to condense, or bundle, the POs in time while maintaining paging capacity, that is, keeping the same number of POs within a given period. This approach ensures that the bundling only affects the distribution of the POs on the timeline, without compromising UE reachability.

As part of this strategy, we are currently proposing configuring additional POs specifically for new (3GPP Rel-19) UEs. These would complement a sparse set of legacy POs (for example, spaced every 160ms), ensuring compatibility with older/legacy UEs. Figure 6 shows a comparison of the legacy PO configuration to a potential Rel-19 configuration including both legacy and the complementary condensed Rel-19 POs.

Figure 6: An exemplary comparison of a legacy paging configuration (top figure, spread POs) and Rel-19 paging configuration (bottom figure) including condensed POs in addition to the legacy POs.

We have evaluated the potential energy-saving gains for several cases, including the new scheme, in comparison to a legacy paging configuration. As shown in Table 1, we observed that a 6 percent to 17 percent energy gain can be achieved (depending on the gap between the POs, that is, PO periodicity) while maintaining the same paging capacity, that is, (the same number of POs per PO cycle).

PCell and PRACH reception: Wisdom is knowing when to listen

Lastly, let’s examine the energy consumption of periodic gNB activity for a PCell from the receiver’s perspective, specifically when listening to PRACH occasions for potential UE access. Today, PRACH reception is performed frequently, even when the cell experiences minimal or no traffic. Our evaluations indicate that even in this case, significant energy savings can be achieved by adapting the frequency of PRACH occasions. For instance, Figure 7 illustrates the energy savings that can be realized by changing the periodicity of PRACH occasions from 10ms to 40ms, as most gNB micro-sleep is replaced with light sleep.

Figure 7: 10ms PRACH occasion periodicity (left) vs. 40ms PRACH occasion periodicity (right).

Similar to the discussion on SSB transmissions, for 5G advanced, we propose that the frequency of PRACH occasions be changed dynamically rather than only semi-statically, as specified today. With the introduction of dynamic adaptation of PRACH occasions in the time domain, we can save energy on the network without impacting initial access performance. Additional PRACH occasions can then be introduced as necessary, as exemplified in Figure 8, using dynamic signaling. This is useful, for example, to avoid congestion caused by PRACH storms during rush hours or if many UEs are paged in conjunction with condensed POs discussed above. This scheme is also backward compatible, as legacy UEs can continue to use the legacy PRACH configuration.

Figure 8. An example of adaptation of PRACH.

Silence is easily broken 

Above, we addressed each of the functions separately, but optimizing them in isolation would not achieve the desired energy savings. Even if one function creates long gaps, another function that wakes up the gNB too frequently can disrupt the gaps needed for our intended energy savings. These functions need to work together cohesively; otherwise, they may be ineffective. Therefore, we need to enable a common design that can flexibly control these adaptations and facilitate the adaptation of features both individually and in conjunction. Below, an example evaluation is shown where both SSB and PRACH occasions are made sparser in time compared to a reference baseline. This saves over three-quarters of gNB energy in low-load scenarios while maintaining the ability to rapidly transition to a higher-load support mode. Compare this to Figure 3 where functions are adapted to various sparse rates individually rather than with a common approach. 

Figure 9: Energy savings comparison of a configuration optimized for performance with 20ms periodicity (left) vs. configuration optimized for energy savings with 160ms periodicity (right).

SCells

A secondary cell (SCell) is a component of CA in cellular networks. It is a secondary connection that can be established in addition to the PCell. SCells are used to increase the data capacity and improve the user experience by allowing for the simultaneous use of multiple component carriers (that is, one or more SCells simultaneously for the same UE), based on current UE traffic needs. The SCell can dynamically be activated or deactivated based on various factors such as buffer occupancy, coverage, and channel quality. 

SCells and SSB transmission: Sparse or dense? Pace it by the case

There is no common signaling on SCells, such as PCell’s paging or random access. They are used solely for data transmission. The only always-on signaling to optimize SCells is related to reference signals, making sure that SCell SSBs do not interrupt gNB sleep. This should make energy optimizations for SCells simpler compared to the challenges faced by PCells.

However, there is an important caveat. SCells are only considered when the capacity of the PCell is less than the current data demand of the UE. Therefore, when SCells are needed, they are needed immediately. If the activation of the SCell takes too long, the data transaction might have already concluded, with the PCell bearing the entire burden, possibly over a longer period than if the SCell had been ready. That’s why SCells are typically kept configured in the background, accompanied by high-rate SSB transmission (For example, every 20ms), to help UEs perform time/frequency synchronization, downlink automatic gain control (AGC), and measurements, ensuring they are ready to be activated immediately on command. Well, guess what—that is exactly what kills every opportunity for gNB to enjoy deeper sleep modes.

Therefore, like PCells, we propose a dynamic approach to SSB provision for SCells as part of the advanced 5G standards, with a slight twist. 

While SCells are kept in the background (configured for the UE but not yet activated), SSBs can either be provided sparsely (for example, every 160ms) as a baseline or not transmitted at all. There is a balance between not transmitting SSBs before activation and transmitting sparsely. Not providing SSBs before SCell activation minimizes gNB energy overhead even further and reduces energy consumption for UEs by not having to measure them. While on the other hand providing SSB enables the gNB to determine which SCells are suitable for a UE. For instance, the gNB may be based on SSB transmission on multiple configured/candidate SCells and depending on received measurement reports from the UE decide which SCells to activate. Although this process of collecting measurement reports and deciding which SCells to activate can create some overhead, the subsequent SCell activation time will be reduced since UEs will have already performed measurements on the SCells to be activated (for example, the SCell may be in a “known” state). These gNB activation decisions depend on factors such as traffic use case, UE type, and the UE’s current coverage or speed conditions. Other factors also play a role in this decision process, such as if other UEs are also currently connected to the same SCell and whether SSBs are already being provided on behalf of other UEs. Perhaps a good compromise lies somewhere in between, including a low SSB provision rate (for example, 160ms) as a baseline for SCell selection which only brings marginal energy cost compared to no SSBs, and activates more SSBs when necessary.

For example, during the SCell activation phase, on-demand SSBs with a very dense pattern (for example, every 5ms to 10ms) can be turned on to ensure rapid activation. Once activated, the pattern can revert to a sparser periodicity (for example, 20ms to 160ms), suitable for the UEs currently connected to that SCell (for example, depending on UE’s speed, coverage, and so on.). A high-level example is illustrated in Figure 10.

Figure 10: Example of dynamic SSB provision for SCell.

In the example above, PCell maintains SSB periodicity while SSB on SCell is transmitted with relevant periodicity on-demand, initially turned off even though the cell is configured for the UE.

We also evaluated the energy savings potential and activation delays associated with various SSB schemes on an SCell. The results are depicted in Figure 11, where different configurations were tested. In all these configurations, the SSB rate of the PCell is every 20ms. When it comes to SCells, in the baseline configuration (shown by the red bar/line in the figure), the SSBs are transmitted with a 20ms periodicity, which is typical in today’s deployments. 

In contrast, another configuration (green bar/line in the figure) implemented a constant SSB rate of 160ms for the SCell, reflecting what energy savings could be achieved according to present specifications. However, as can be seen, this setup resulted in increased latency and was not a practical configuration.

Lastly, we explored the newly discussed functionality for NR advanced where the SCell initially transmits SSBs at a 160ms periodicity (blue bar/line in the figure), switching to a 10ms periodicity during activation, and back to 160ms afterward to maintain the connection. This approach demonstrates significant energy savings without considerably impacting packet scheduling delays – exactly the type of solution we set out to find at the beginning of this post.

Figure 11: Evaluation for setups including a PCell and an SCell. Comparing the energy cost to the corresponding configuration's packet scheduling delay.

As energy costs rise and mobile network demands escalate, energy consumption and the cost of operating the networks are also increasing. The evaluations above highlight the potential for energy savings in different time-domain transmission schemes within 5G Advanced networks, demonstrating negligible performance impact. As discussed, even individually small and seemingly insignificant transmissions accumulate leading to unnecessary energy consumption within mobile networks. These findings indicate that always-on transmission schemes are not necessary for 5G systems. We need to keep this in mind as we continue our work in 5G Advanced and ensure that this mindset is integrated beyond the standards of 5G. 

Read more

• Explore our reports on Network energy performance 
• Learn about energy efficiency in Ericsson’s 5G RAN 
• Breaking the energy curve: Network energy consumption modeling and energy saving technologies.