Skip navigation

Achieving Gigabit connectivity with Wi-Fi today

The Digital Decade Policy Programme aims for Europe’s digital transformation by 2030, setting, among others, the Gigabit connectivity target for fixed broadband connections delivered to the home. Wi-Fi plays a key role in maximizing the capacity that reaches the consumer. Efforts should be put on optimizing Wi-Fi operation in dense scenarios with appropriate channel bandwidth and features, rather than overprovisioning of spectrum.

Executive summary

The Digital Decade Policy Programme (DDPP) sets concrete targets and objectives for 2030 to guide Europe’s digital transformation. Digital infrastructure is one of the cardinal points focusing on improving and developing mobile and fiber network infrastructures. The goal is to ensure Gigabit connectivity for everyone and achieve high-speed mobile coverage (at least 5G) everywhere” [1]. The Commission sets the Gigabit connectivity target for fixed broadband connections delivered to the home [2]. Fixed broadband connectivity is typically delivered to households via fiber or fixed wireless access. This is then connected to a Wi-Fi router, which distributes the connection within the home/enterprise. Wi-Fi plays a key role in maximizing the capacity that reaches the consumer.

There are diverging views on how to prevent Wi-Fi from becoming a bottleneck in delivering Gigabit connectivity to consumers. Some request additional spectrum in the upper 6 GHz band [3] while others point to efficiency deficits and argue for more efficient usage of the current spectrum [4].

We have analyzed Wi-Fi performance in an apartment building, considering the most common channel allocation in Europe, namely the 5 GHz and lower 6 GHz bands, for different channel bandwidths, and concluded that:

  • Speeds significantly higher than 1 Gbps can be achieved today with current Wi-Fi technology.
  • The best performance is achieved when combining efficient reuse of the available channels with modern Wi-Fi features.
  • Larger channel bandwidth does not always mean better performance. Throughput increases, but interference among access points (APs) increases as well.

Further emphasis should be put on optimizing operation in dense scenarios with appropriate channel bandwidth and features, rather than overprovisioning of spectrum.

Spectrum availability

Historically, Europe has had two spectrum ranges where Wi-Fi technologies can be deployed: 2.4 GHz, 2400-2483.5 GHz [5] and 5 GHz, 5150-5350 MHz and 5470-5725 MHz [6].

Figure 1: 2.4 GHz (top) and 5 GHz (bottom) European license-exempt allocations and channel bandwidths supported by the ETSI standard [7], [8]
Figure 1: 2.4 GHz (top) and 5 GHz (bottom) European license-exempt allocations and channel bandwidths supported by the ETSI standard [7], [8]

Figure 1: 2.4 GHz (top) and 5 GHz (bottom) European license-exempt allocations and channel bandwidths supported by the ETSI standard [7], [8]

In November 2020, Europe decided to nearly double the amount of spectrum allocated for RLAN by making the lower 6 GHz band available for license-exempt use, 5945-6425 [9].

Figure 2: Lower 6 GHz European license-exempt allocations and channel bandwidths supported by the ETSI standard [10].

Figure 2: Lower 6 GHz European license-exempt allocations and channel bandwidths supported by the ETSI standard [10].

Wi-Fi standards evolution

Wi-Fi is the most common technology being deployed under “license-exempt” authorization regimes, and its operation in these bands is at the core of the underlying standard, IEEE 802.11.

In 1997, the first version of the standard [11] supported up to 2 Mb/s throughput using infrared or radio transmissions in the 2.4 GHz band. In 1999, IEEE 802.11a rapidly incorporated the use of the newly available license-exempt 5 GHz band, supporting data rates of up to 54 Mb/s. Accessing a channel from the begining of the standards is fundamentally based on the principle that a user first senses if someone else is transmitting before starting its own transmission. If someone else is transmitting, the user defers. This method is known as carrier-sense multiple access with collision avoidance (CSMA/CA). The sensing duration is random to avoid simultaneous transmissions by multiple users (known as collisions). CSMA/CA, by design, is completely distributed, eliminating the need for central coordination. It is therefore a good match for the license-exempt spectrum, where no central authority or ownership exists.

A significant improvement was witnessed in 2009, when the amendment "n" (known as Wi-Fi 4) introduced multi-antenna transmissions and the doubling of the channel bandwidth to 40 MHz, increasing the maximum raw data rate to 600 Mb/s. At this rate, a transmission of 1000 B takes about 126 µs (14.4μs for the transmission of 4 OFDM symbols containing the data, plus 52μs for the preamble, 16μs inter-frame space, and 44μs for the acknowledgment). However, the channel access procedure on average roughly doubles this duration to 236 µs. This large overhead for every frame significantly results in an achieved throughput of only 33.4 Mb/s in this example – an efficiency of ~5 percent. Improving the channel access procedure itself would have introduced fairness problems with the large installed base of legacy users using the spectrum. Hence, the new generation instead added the capability to aggregate multiple small frames into one longer transmission. In this way, a single channel access is sufficient to transmit and acknowledge multiple frames, reducing overheads per frame significantly.

This principle continues in the subsequent standard amendment "ac" (known as Wi-Fi 5): the raw data rate is increased by quadrupling the maximum channel bandwidth to 160 MHz and stuffing more bits into one transmitted symbol, allowing a maximum raw data rate of 7 Gb/s – resulting in an even lower efficiency for small frames as the channel access procedure before the transmission again remained unchanged. As in the previous amendment, this is mitigated by adding more options to aggregate multiple frames per channel access. For Wi-Fi 5, the strategy is to add multi-user multiple-input multiple-output (MU-MIMO), allowing it to serve up to four users with one channel access.

The amendment "ax" or Wi-Fi 6 then introduced orthogonal frequency-division multiple access (OFDMA), which allows transmissions to or from up to 74 users at the same time by assigning different frequency parts of the current channel bandwidth to them. Surprisingly, this amendment was the first of the major amendments that did not enlarge the channel bandwidth. However, it was the first time since 1999 that a substantial amount of spectrum was added, the 6 GHz band. In the market, these products are certified by Wi-Fi Alliance (WFA) as Wi-Fi 6E.

The latest Wi-Fi generation, Wi-Fi 7, based on the amendment "be", further enlarges the maximum channel bandwidth to 320 MHz, resulting in a maximum raw data rate of up to 23 Gb/s. The improvements in aggregation were achieved through the ability to parallelize multiple channel access procedures at the same time. This allowed for simultaneous transmission and reception across multiple channels and is referred to as "multi-link operation" in the standard.

This overview of Wi-Fi's generations offers a simplified perspective on its developments as it does not account for the numerous other minor amendments that have been introduced since 1997. However, it is sufficient to show the general trend of the development: Wider channel bandwidths (from 20 MHz in the beginning to now 320 MHz) increase the raw data rate; features like frame aggregation, multi-user transmissions, and multi-link operation are employed to retain sufficient efficiency and thus throughput.

Although the new features that came with each generation are impressive, the basic channel access mechanism based on CSMA/CA remains untouched and thus is still the bottleneck for Wi-Fi. The impact of this bottleneck depends on how efficiency-increasing features perform under realistic conditions.

Detailed analysis: Today's Wi-Fi offers Gigabit speeds through 5 GHz and lower 6 GHz bands

In the following sections, we analyze what speeds Wi-Fi can deliver today, deploying spectrum in the 5 GHz and lower 6 GHz bands.

Model scenario

We conduct system-level simulations of a multi-user, multi-cell Wi-Fi installation in an apartment building. In particular, we simulate a smaller version of the residential scenario described in [12], which is also used in previous studies [13], [3].

Figure 3: A simple depiction of the 2x5x3 residential building

Figure 3: A simple depiction of the 2x5x3 residential building. Source: MATLAB tools

The residential scenario consists of a building of three floors with 10 apartments per floor, resulting in a total of 30 apartments in the building. Each apartment is 10 m by 10 m in size and 3 m high. Within each apartment, one Wi-Fi access point (AP) and one user are placed randomly. We assume that there is no speed limitation in terms of the fixed connection delivered to the APs in the apartment and that the Wi-Fi connection to the user is the factor that limits throughput.

Channel assignment

We assume the most common channel allocation in Europe for two-channel bandwidths, 80 MHz and 160 MHz:

  • 11 channels with 1x80 MHz bandwidth: Five at 5 GHz and six at lower 6 GHz. This corresponds to a total of 880 MHz.
  • Five channels with 1x160 MHz bandwidth: Two at 5 GHz and three at lower 6 GHz. This corresponds to a total of 800 MHz.

We also consider the latest commercially deployed technology Wi-Fi 7 and therefore available features, such as multi-link operation (MLO) in accordance with the Wi-Fi alliance (WFA) certification requirements for Wi-Fi 7. For example from TP-Link [16], Netgear [17], Asus [18] for the home environment/endconsumer market. This was also suggested by [4]. This means that a single AP can use multiple channels concurrently within each apartment to communicate to its users:

  • Five channels 2x80 MHz: Five channels of 80 MHz, using MLO with two links per AP. This corresponds to a total of 800 MHz.
  • Three channels 3x80 MHz: Three channels of 80 MHz, using MLO with three links per AP. This corresponds to a total of 720 MHz.

All considered assignments roughly use the same amount of spectrum, ranging from 720 MHz to 880 MHz; however, they differ in how this spectrum is allocated among the apartments. Allocating more spectrum per apartment increases the potential capacity, but also decreases the number of available channels and thus the distance between two APs using the same channel, known as the "reuse distance", which increases interference.

For all channel and multi-link configurations, we distribute the available channels across the APs in the apartment building. We assume a reasonably good channel selection algorithm that maximizes the distance between two APs using the same channels, thereby minimizing inter-AP interference. Figures 4-6 show simple depictions of how the channels are distributed across the building for the case when there are three, five, and 11 channels to share respectively. A red box indicates that the apartments share the same channels while a grey box indicates that they use another channel. In the scenarios below, 10, six, and three apartments need to share the same channel for three, five, and 11 channels, respectively.

Figure 4: Channel allocation for the three-channel configuration

Figure 4: Channel allocation for the three-channel configuration. Source: MATLAB tools

Figure 5. Channel allocation for the five-channel configuration

Figure 5. Channel allocation for the five-channel configuration. Source: MATLAB tools

Figure 6. Channel allocation for the eleven-channel configuration

Figure 6. Channel allocation for the eleven-channel configuration. Source: MATLAB tools

Simulation methodology

The simulator used in this study is a full-fledged system simulator featuring detailed models of the Wi-Fi physical layer and the medium access control (MAC) layer.

The simulator employs an event-driven approach, which means that it simulates not only the start and end of each transmission during the simulation time, but also all details of Wi- Fi's carrier-sense multiple access with collision avoidance (CSMA/CA)-based channel access procedure. This means that a node (AP or user) will determine if the channel is occupied before each transmission using both a preamble and energy detection. It defers from transmission until the channel becomes idle.

Implementing the channel access procedure becomes crucial in interference-limited scenarios, as it leads to substantially different spectrum usage compared to a simulation that only considers the physical layer and positional variances, as in [13], [3].

Furthermore, each node selects the link rate of each transmission using the Minstrel link adaptation algorithm [14] based on recent packet error rate statistics. At the receiver side, the current signal-to-noise ratio (SNR) is mapped to an error probability, which depends on the chosen data rate, with higher rates requiring a higher SNR. The error probability sets the threshold for a random number between 0 and 1 that determines if the packet is received successfully or must be retransmitted.

We assume two different types of traffic as load offered to the system: Downlink full buffer and File Transfer Protocol (FTP )downloads. As the name suggests, in the full buffer case, the transmit buffer at every AP is always full. Hence, during the simulation period, an AP is either transmitting or waiting for the channel to become available to begin transmission. There are no idle periods since the load is maximized. Although the full-buffer assumption offers insights into an extreme case, it has often been criticized for being unrealistic as it allows unreasonable aggregation lengths and thus skews the results for the later Wi-Fi generations. Thus, we also modeled FTP file downloads by adding a full FTP/TCP/IP (File Transfer Protocol/Transmission Control Protocol/Internet Protocol) implementation on top of the Wi-Fi stack. Here, the transmission buffer level is controlled by TCP and is not arbitrarily large. As before, no idle times occur because there is no gap between the completion of one FTP file download and the start of the next one.

A final notable assumption is the antenna configuration of the APs and the users. To achieve the advertised maximum data rates of the latest Wi-Fi generation, multiple antennas are used to send multiple data streams in parallel. Wi-Fi 7 allows up to eight streams, which requires the AP and the user device to have eight antennas each, and results in a maximum raw data rate of ~5.6 Gb/s on an 80 MHz channel and ~10.4 Gb/s on a 160 MHz channel. However, due to size and complexity limitations, Wi-Fi 7 APs on the market are equipped with a maximum of four antennas per band. Similarly, only a few user devices have more than two antennas. Therefore, our simulations assume four and two antennas for the AP and user, respectively. This limits the maximum raw data rate to ~1.4 Gb/s and ~2.8 Gb/s on an 80 MHz channel and a 160 MHz channel, respectively.

The essential parameters for the simulations are summarized in Table 1.

Table 1: Simulation parameters

Simulation parameter Value
Antenna configuration AP: 4 Tx 4 Rx per link
User: 2 Tx 2 Rx per link
Transmit power 23 dBm
Pathloss model according to [12], where
f is the center frequency,
d is the link distance,
F is the number of floors traversed,
W is the number of walls traversed in x-direction plus the number of walls traversed in y-direction,
w is loss per wall
pathloss(d) = 40.05
+ 20*log10(f/2.4)
+ 20*log10(min(d,5))
+ (d>5) * 35*log10(d/5)
+ 18.3*F^((F+2)/(F+1)-0.46)
+ w*W
Wall loss 5 to 18 dB
Link adaptation algorithm Minstrel [14]
Maximum modulation and coding scheme (MCS) and streams 4096-QAM 5/6, 2 streams
(corresponding to ~2.8Gb/s for 160 MHz maximum raw data rate) pathloss(d) = 40.05
Traffic load
  1. Full buffer downlink traffic
  2. Consecutive FTP download via TCP/IP
Key performance indicator Delivered throughput measured at the user
Drops (random locations of the APs and users in the apartment) per configuration 100

Results

Figure 7 shows the mean downlink throughput across all simulated users. We vary the path loss induced by each wall: A low value represents drywall, while a high value represents concrete walls. This enables us to model different types of buildings, such as modern and old buildings. The different lines correspond to the channel configurations described earlier.

Figure 7. Mean downlink throughput across different wall loss values for the channel configurations

Figure 7. Mean downlink throughput across different wall loss values for the channel configurations. Source: MATLAB tools

We can see that all configurations can achieve more than 1 Gb/s in mean throughput.

In general, as the attenuation of the walls increases, so does achievable throughput. This is due to the wall becoming more effective at blocking interference between the apartments, providing better spatial reuse. This effect cannot be seen in the case of 11 channels 1x80 MHz configurations, as the distance between APs using the same channel is sufficient to limit the inter-AP interference even for the 5 dB wall loss. Thus, for any realistic wall loss (for example 11 dB for an inter-apartment wall as in [3]) we can infer that this assignment is not interference-limited and does not use the available spectrum efficiently, as it uses more channels than necessary to provide a large reuse distance.

Comparing the cases of 11 channels of 1x80 MHz and five channels of 1x160 MHz, we observe that:

  • A smaller number of wider channels increases inter-AP interference and thus lowers achievable throughput, but
  • At the same time, it also gives more bandwidth per apartment, which increases the achievable throughput.

The second effect dominates, increasing the throughput from 1.3 Gb/s to 2.2 Gb/s, with greater gains observed for higher wall losses. This shows that Wi-Fi can take advantage of the opportunities for spatial reuse provided by increased attenuation.

If we consider enabling the multi-link operation in Wi-Fi 7 and allowing 2 x 80 MHz per AP, we observe an increase in delivered throughput compared to using 160 MHz channels. This demonstrates how parallel use of two channels increases efficiency.

Further enhancing multi-link operation to 3 x 80 MHz links per AP has two opposing effects: On the one hand, using three links in parallel increases the possible throughput in comparison to the two-link case. On the other hand, inter-AP interference per channel is increased as the assignment of three channels per AP results in a smaller reuse distance. The results indicate that even at low wall loss, the first effect is already dominant. This becomes even more visible with higher wall loss settings as they offer more options for spatial reuse.

The average values shown in the figure above cannot fully capture the complete picture as they conceal the distribution of the values. Hence, to get a deeper understanding of how throughput is experienced at all APs, we also plotted the cumulative distribution function (CDF) of the throughput of the APs for the 11 dB wall loss setting.

Figure 8. CDF of the downlink throughput of all APs sharing the same channel

Figure 8. CDF of the downlink throughput of all APs sharing the same channel. Source: MATLAB tools

In the CDF, we can see every throughput value for all APs and simulations recorded.

It can be observed that the CDF of the 11-channel configuration's CDF is essentially a straight line. All APs have the same throughput, independent of the location of the apartment or their positioning inside the apartment. This confirms our earlier conclusion that for this configuration, the APs are not affected by any interference as there is overprovisioning of the number of channels. Five channels of 160 MHz not only increase the mean throughput but also the variance and maximum throughput, compared to 11 channels of 80 MHz. About 50 percent of the measured throughput samples are twice as high as the maximum for the 11 x 80 MHz case, and 93 percent achieve a higher throughput than it.

Adding multi-link operation to the AP shows further throughput gains, but also larger variance, especially for the three-link case with the smallest channel reuse distance. Here, the maximum throughput (almost 4 Gb/s) is achieved in less than 20 percent of the samples, but overall there is higher throughput compared to the other options.

Table: Mean and the 5-/95-percentile throughput achieved with the different channel configurations for a wall loss of 11 dB

Channel configuration Mean throughput for 11 dB wall loss 95 percent throughput for 11 dB wall loss 5 percent throughput for 11 dB wall loss
11 channels of 80 MHz (non interference-limited) 1.31 Gb/s 1.32 Gb/s 1.21 Gb/s
Five channels of 160 MHz 2.13 Gb/s 2.44 Gb/s 1. 10 Gb/s
Five channels of 80 MHz, using two links per AP 2.27 Gb/s 2.64 Gb/s 0.76 Gb/s
Three channels of 80 MHz, using three links per AP 2.59 Gb/s 3.94 Gb/s 0.95 Gb/s

For all three interference-limited configurations, we observe a wide range of values, particularly when compared to the 11 channels of the 80 MHz configuration. While their mean is well above 1 Gb/s, their five-percentiles (the worst five percent of the samples) are around 1 Gb/s. The wide range of values is due to the short-term unfairness of the CSMA/ CA medium access: When a collision occurs, the AP increases its window for the random variable that determines the sensing duration. This lowers the probability of another collision. Due to the assumption of full buffer traffic, nearby nodes are continuously using the medium, causing the collided AP to take a long time to achieve its next successful transmission. This results in long delays and reduced throughput. This unfairness occurs more often to the APs in the center of the building, as they have more neighbors than the ones on the edges of the residential building. This effect is well known in the literature [15], but still persists in the latest Wi-Fi generation. In the 11-channel configuration, it does not occur due to overprovisioning of channels.

While a full buffer traffic model is simple to simulate and understand, it does not accurately reflect reality because no application is designed to transmit data continuously. Thus, we also simulate an FTP download traffic model where a user requests an infinite number of file downloads via a TCP connection, one after the other. As such, this is also a very high load scenario, but it includes realistic behavior of the transport and IP layers, thus more accurately reflecting a real application. Figure 9 compares the throughput CDFs of the full buffer traffic model and the FTP traffic model, both configured with a path loss of 11 dB through walls.

Figure 9. A comparison of the CDF of the downlink throughput at 11 dB wall loss between the full buffer and FTP traffic models for all channel configurations

Figure 9. A comparison of the CDF of the downlink throughput at 11 dB wall loss between the full buffer and FTP traffic models for all channel configurations. Source: MATLAB tools

As expected, the overhead of the FTP, but also the congestion control of the TCP and the resulting reduced efficiency due to less aggressive packet aggregation reduces the achieved throughput by roughly 10-20 percent. Hence, our conclusions from above still hold even under more realistic traffic assumptions. In particular, we note that with more realistic traffic, over 1 Gb/s is, in general, achieved in 99 percent of the cases.

As a final metric, we calculate the user spectral efficiency under the assumption of the full buffer traffic model, given by the mean user throughput divided by the number of channels and their bandwidth. Figure 10 shows the spectral efficiency across all four considered configurations, providing additional insight into how effectively Wi-Fi utilizes the available spectrum under a dense deployment with 3 links per AP and 80 MHz channels. In comparison, using 11 separate channels results in achieving less than half the spectral efficiency.

Figure 10. Mean spectral efficiency

Figure 10. Mean spectral efficiency. Source: MATLAB tools

Conclusion

Wi-Fi has tremendously improved the speed of raw data rates across generations, from 54 Mb/s to 23 Gb/s. To achieve these rates, Wi-Fi has been increasing its channel bandwidth as well as adding features to increase the raw data rate.

In this study, we analyzed how Wi-Fi continues to successfully convert raw numbers into real and tangible user benefits. It also examines the data rates users can achieve today using the available spectrum for license-exempt use.

Based on our results, the best performance is achieved by combining an efficient reuse of the available channels with modern Wi-Fi features such as multi-link operation. Additional features such as BSS-coloring to improve spatial reuse (Wi-Fi 6) or inter-AP coordination (under discussion for Wi-Fi 8), would demonstrate further gains. However, these features were not considered in this study.

Still, the limiting factor in dense channel reuse scenarios is the channel access mechanism, CSMA/CA, which has not changed since the first generation of Wi-Fi. It leads to the penalization of nodes in sensitive positions under high loads, and thus to an unfair distribution of throughput. With the current feature set, this can only be solved by a severe overprovisioning of the system with respect to spectrum usage, unnecessarily lowering spectral efficiency.

In summary, Wi-Fi can reliably deliver speeds significantly exceeding 1 Gb/s today under realistic conditions by combining the available spectrum in the 5 GHz and lower 6 GHz bands. Rather than encouraging overprovisioning of spectrum, more emphasis should be put on optimizing operations in dense scenarios with appropriate channel bandwidths and features.

Contributors

Charlie Pettersson

Charlie Pettersson is a senior researcher at Ericsson with extensive experience in wireless communication technologies. He has been actively participating on the IEEE 802.11 standardization efforts since 2018. Charlie specializes in modelling radio networks for performance optimization. With deep technical expertise and a passion for innovation, he works to shape the future of connectivity by enhancing the efficiency and reliability of wireless networks.

Sebastian Max

Sebastian Max is a master researcher at Ericsson Research. Since 2005, Sebastian has been actively participating in various standards development organizations, including IEEE 802.11/.15/.18, Bluetooth SIG, and ETSI BRAN. He received his Dr.-Ing. degree for his thesis on IEEE 802.11s mesh networking in 2011 and joined Ericsson in 2018. Currently he serves as the Vice-Chair of the IEEE 802.11 Coexistence Standing Committee and as the Secretary of the IEEE 802.11 Task Group bp (Ambient Power Communication). His research focuses on performance evaluation of license-exempt radio technologies and multi- standard spectrum sharing.

Erika Tejedor

Erika Tejedor is Vice President of spectrum regulations at Ericsson and focuses on ensuring harmonized future spectrum availability and favourable regulations for Ericsson businesses. During her career, she has worked across research, product development, standardization, industry partnerships and spectrum regulations. Erika graduated from the University of Zaragoza (Spain) and the Linköping University (Sweden) and holds a Master in Electrical and Electronics Engineering as well as in Wireless Communications.

References

2
Decision (EU) 2022/2481
11
"IEEE standard for wireless LAN medium access control (MAC) and physical layer (PHY) specifications," in IEEE Std 802.11-1997 , November 1997, doi: 10.1109/ IEEESTD.1997.85951.
14
D. Xia, J. Hart and Q. Fu, "Evaluation of the Minstrel rate adaptation algorithm in IEEE 802.11g WLANs," 2013 IEEE International Conference on Communications (ICC), Budapest, Hungary, 2013, pp. 2223-2228, doi:10.1109/ICC.2013.6654858.