Problems and Requirements of Satellite Constellation for Internet

5.2.1. Inter-satellite Communication Overview

In order to form a network by satellites, there must be an inter-satellite communication. Traditionally, inter-satellite communication uses the microwave technology, but it has following disadvantages:¶

1. Bandwidth is limited and only up to 600M bps [Microwave-vs-Laser-communication].¶
2. Security is a concern since the microwave beam is relatively wide and it is easy for 3rd party to sniff or attack.¶
3. Big antenna size.¶
4. Power consumption is high.¶
5. High cost per bps.¶

Recently, laser is used for the inter-satellite communication, it has following advantages, and will be the future for inter-satellite communication.¶

1. Higher bandwidth and can be up to 10G bps [Microwave-vs-Laser-communication].¶
2. Better security since the laser beam size is much narrower than microwave, it is harder for sniffing.¶
3. The size of optical lens for laser is much smaller than microwave's antenna size.¶
4. Power saving compared with microwave.¶
5. Lower cost per bps.¶

The range for satellite-to-satellite communications has been estimated to be approximately 2,000 km currently [Laser-communication-range].¶

From Table 2, we can see the Space Distance (SD) for some LEO (altitude over 1100km) are exceeding the celling of the range of laser communication, so, the satellite and orbit density for LEO need to be higher than the estimation values in the Table 1.¶

Assume the laser communication is used for inter-satellite communication, then we can analyze the lifetime of inter-satellite communication when satellites are moving. The Figure 3 illustrates the movement and relative position of satellites on three orbits. The inclination of orbit planes is 90 degrees.¶

                       + North pole
                      /|\
                     | s |
                    s  |  s
                   /   s   \
                   s   |   s
                   |   s1  |
                   s4  |   s6
                   |   s2  |   -------- Equator
                   s5  |   s7
                   |   s3  |
                   s   |   s
                   \   s   /
                    s  |  s
                     | s |
                      \|/
                       +  South pole

There are four scenarios:¶

1. For satellites within the same orbit

The satellites in the same orbit will move to the same direction with the same speed, thus the interval between satellites is relatively steady. Each satellite can communicate with its front and back neighbor satellite as long as satellite's orbit is maintained in its life cycle. For example, in Figure 3, s2 can communication with s1 and s3.¶

2. For satellites between neighbor orbits in the same group at non-polar areas

The orbits for the same group will share the same orbit altitude and inclination. So, the satellite speed in different orbit are also same, but the moving direction may be same or different. Figure 4 illustrates this scenario. When the moving direction is the same, it is similar to the scenario 1, the relative position of satellites in different orbit are relatively steady as long as satellite's orbit is maintained in its life cycle. When the moving direction is different, the relative position of satellites in different orbit are un-steady, this scenario will be analyzed in more details in Section 5.2.2.¶

3. For satellites between neighbor orbits in the same group at polar areas

For satellites between neighbor orbits with the same speed and moving direction, the relative position is steady as described in #2 above, but the steady position is only valid at areas other than polar area. When satellites meet in the polar area, the relative position will change dramatically. Figure 5 shows two satellites meet in polar area and their ISL facing will be swapped. So, if the range of laser pointing angle is 360 degrees and tracking technology supports, the ISL will not be flipping after passing polar area; Otherwise, the link will be flipping and inter-satellite communication will be interrupted.¶

4. For satellites between different orbits in the different group

The orbits for the different group will have different orbit altitude, inclination and speed. So, the relative position of satellite is not static. The inter-satellite communication can only last for a while when the distance between two satellite is within the limit of inter-satellite communication, that is 2000km for laser [Laser-communication-range], this scenario will be analyzed in more details in Section 5.2.3¶

            i+N/2         i+1+N/2       i+2+N/2
          / \           / \           / \
         /   \         /   \         /   \
        S1    \       S2    \       S3    \
       /       S4    /      S5     /      S6
      /         \   /         \   /         \
     /           \ /           \ /           \
     i-1           i             i+1

* The total number of orbit planes are N
* The number (i-1, i, i+1,...) represents the Orbit index
* The bottom numbers (i-1, i, i+1) are for orbit planes on
  which satellites (S1, S2, S3) are moving from bottom to up.
* The top numbers (i+N/2, i+1+N/2, i+2+N/2) are for orbit
  planes on which satellites (S4, S5, S6) are moving from up
  to bottom.

             \      /
              P3   P4
               \  /
                \/
                /\
               /  \
              P1   P2
             /      \

* Two satellites S1 and S2 are at position P1 and P2 at time T1
* S1's right facing ISL connected to S2's left facing ISL
* S1 and S2 move to the position P4 and P3 at time T2
* S1's left facing ISL connected to S2's right facing ISL

5.2.2. Satellites on Adjacent Orbit Planes with Same Altitude

For satellites on different orbit planes with same altitude, the estimation of the lifetime when two satellite can communicate are as follows.¶

Figure 6 illustrates a general case that two satellites move and intersect with an angle A.¶

                  ^ V2
                 /
                /
               +-
              /  \ A
-------------+----+----> V1
            /
           /

More specifically, for orbit planes with the inclination angle i, Figure 7 illustrates two satellites move in the opposite direction and intersect with an angle 2*i.¶

                     ^ move from south to north
             \      /
              \    /
               \  /\
                \/  | A = 2*i
                /\  |
               /  \/
              /    \
             /      V move from north to south

Follows are the math to calculate the lifetime of communication. Table 3 are the results using the math for two satellites with different altitudes and different inclination angles.¶

Dl

The laser communication limit, Dl=2000km [Laser-communication-range]¶

A

The angle between two orbit's vertical projection on Earth. A=2*i¶

V1

The speed vector of satellite on orbit1¶

V2

The speed vector of satellite on orbit2¶

|V|

the magnitude of the difference of two speed vector V1 and V2, |V|=|V1-V2|=sqrt((V1-V2*cos(A))^2+(V2*sin(A))^2). For satellites with the same altitude and inclination angle i, V1=V2, so, |V|=V1*sqrt(2-2*cos(2*i))=2V1*sin(i)¶

T

The lifetime two satellites can communicate, or the time of two satellites' distance is within the range of communication, T = 2*Dl/|V|.¶

5.2.3. Satellites on Adjacent Orbit Planes with Different Altitude

For satellites on different orbit planes with different altitude, the estimation of the lifetime when two satellite can communicate are as follows.¶

Figure 8 illustrates two satellites (with the altitude difference Da) move and intersect with an angle A.¶

                ^ V2
               /
              /
      -------+  /
      Da    /| +-
           / |/  \ A
----------/--+----+----> V1
         /  /
           /
          /
         /

Follows are the math to calculate the lifetime of communication¶

Dl

The laser communication limit, Dl=2000km [Laser-communication-range]¶

Da

Altitude difference (in km) for two orbit planes¶

A

The angle between two orbit's vertical projection on Earth¶

Vl

The speed vector of satellite on orbit 1¶

V2

The speed vector of satellite on orbit 2¶

|V|

the magnitude of the difference of two speed vector V1 and V2, |v|=|V1-V2|=sqrt((V1-V2*cos(A))^2+(V2*sin(A))^2)¶

T

The lifetime two satellites can communicate, or the time of two satellites' distance is within the range of communication, T = 2*sqrt(Dl^2-Da^2)/|V|¶

Using formulas above, below is the estimation for the life of communication of two satellites when they intersect. Table 4 and Table 5 are for two VLEOs with the difference of 114.1km for altitude. (VLEO1 and VLEO2 on Table 2). Table 6 and Table 7 are for two LEOs with the difference of 175km for altitude (LEO2 and LEO3 on Table 2).¶

7.2.1. One Satellite Relay

One satellite relay is the simplest method for satellite constellation to provide Internet service. By this method, IP traffic will be relayed by one satellite to reach the DGS and go to Internet.¶

The solution option and associated requirements are:¶

S1. The satellite only does L1 relay or the physical signal process.¶

For this solution, a satellite only receives physical signal, amplify it and broadcast to ground stations. It has no further process for packet, such as L2 packet compositing and processing, etc. All packet level work is done only at ground station. The requirements for the solution are:¶

R1-1. SGS and BGS are configured as IP routing node. Routing protocol is running in SGS and BGS

SGS and BGS is a IP peer for a routing protocol (IGP or BGP). SGS will send internet traffic to DGS as next hop through satellite uplink and downlink.¶

R1-2. DGS must be connected with Internet.

DGS can process received packet from satellite and forward the packet to the destination in Internet.¶

In addition to the above requirements, following problem should be solved:¶

P1-1. IP continuity between two ground stations

This problem is that two ground stations are connected by one satellite relay. Since the satellite is moving, the IP continuity between ground stations is interrupted by satellite changing periodically. Even though this is not killing problem from the view point that IP service traditionally is only a best effort service, it will benefit the service if the problem can be solved. Different approaches may exist, such as using hands off protocols, multipath solutions, etc.¶

S2. The satellite does the L2 relay or L2 packet process.¶

For this solution, IP packet is passing through individual satellite as an L2 capable device. Unlike in the solution S1, satellite knows which ground station it should send based on packet's destination MAC address after L2 processing. The advantage of this solution over S1 is it can use narrower beam to communicate with DGS and get higher bandwidth and better security. The requirements for the solution are:¶

R2-1. Satellite must have L2 bridge or switch capability

In order to forward packet to properly, satellite should run some L2 process such as MAC learning, MAC switching. The protocol running on satellite must consider the fast movement of satellite and its impact to protocol convergence, timer configuration, table refreshment, etc.¶

R2-2. same as R1-1 in S1

R2-3. same as R1-2 in S1

In addition to the above requirements, the problem P1-1 for S1 should also apply.¶

7.2.2. Multiple Satellite Relay

For this method, packet from SGS will be relayed through multiple intermediate satellites and ground station until reaching a DGS.¶

This is more complicated than one satellite relay described in Section 7.2.1.¶

One general solution is to configure both satellites and ground-stations as IP routing nodes, proper routing protocols are running in this network. The routing protocol will dynamically determine forwarding path. The obvious challenge for this solution is that all links between satellite and ground station are not static, according to the analysis in Section 5.1, the lifetime of each link may last only couple of minutes. This will result in very quick and constant topology changes in both link state and IP adjacency, it will cause the distributed routing algorithms may never converge. So this solution is not feasible.¶

Another plausible solution is to specify path statically. The path is composed of a serials of intermediate ground stations plus SGS and DGS. This idea will make ground stations static and leave the satellites dynamic. It will reduce the fluctuation of network path, thus provide more steady service. One variant for the solution is whether the intermediate ground stations are connected to Internet. Separated discussion is as below:¶

S1. Manual configuring routing path and table¶

For this solution, the intermediate ground stations and DGS are specified and configured manually during the stage of network planning and provisioning. Following requirements apply:¶

R1-1. Specify a path from SGS to DGS via a list of intermediate ground stations.

The specified DGS must be connected with internet. Other specified intermediate ground stations does not have to¶

R1-2. All Ground stations are configured as IP routing node.

Static routing table on all ground stations must be pre-configured, the next hop of routes to Internet destination in any ground station is configured to going through uplink of satellite to the next ground station until reaching the DGS.¶

R1-3. All Satellites are configured as either L1 relay or L2 relay.

The Satellite can be configured as L1 relay or L2 relay described in S1 and S2 respectively in Section 7.2.1¶

In addition to the above requirements, the problem P1-1 in Section 7.2.1 should also apply.¶

S2. Automatic decision by routing protocol.¶

This solution is only feasible after the IP continuity problem (P1-1 in Section 7.2.1) is solved. Following requirements apply:¶

R2-1. All Ground stations are configured as IP routing node. Proper routing protocols are configured as well.

The satellite link cost is configured to be lower than the ground link. In such a way, the next hop of routes for the IP forwarding to Internet destination in any ground station will be always going through the uplink of satellite to the next ground station until reaching the DGS.¶

R2-2. All Satellites are configured as either L1 relay or L2 relay.

The Satellite can be configured as L1 relay or L2 relay described in S1 and S2 respectively in Section 7.2.1¶

In addition to the above requirements, the problem P1-1 in Section 7.2.1 should also apply.¶

7.3.1. L2 or L3 network

The 1st question to answer is should the satellite network be configured as L2 or L3 network? As analyzed in Section 4.2 and Section 4.3, since there are couple of hundred or over ten thousand satellites in a network, L2 network is not a good choice, instead, L3 or IP network is more appropriate for such scale of network.¶

7.3.2. Inter-satellite-Link Lifetime

If we assume the orbit is circular and ignore other trivial factors, the satellite speed is approximately determined by the orbit altitude as described in the Section 5.1. The satellite orbit can determine if the dynamic position of two satellites is within the range of the inter-satellite communication. That is 2000km for laser communication [Laser-communication-range] by Inter Satellite Laser Link (ISLL).¶

When two satellites' orbit planes belong to the same group, or two orbit planes share the same altitude and inclination, and when the satellites move in the same direction, the relative positions of two satellites are relatively stationary, and the inter-satellite communication is steady. But when the satellites move in the opposite direction, the relative positions of two satellites are not stationary, the communication lifetime is couple of minutes. The Section 5.2.2 has analyzed the scenario.¶

When two satellites' orbit planes belong to the different group, or two orbit planes have different altitude, the relative position of two satellite are unstable, and the inter-satellite communication is not steady. As described in Section 5.2, The life of communication for two satellites depends on the following parameters of two satellites:¶

1. The speed vectors.¶
2. The altitude difference¶
3. The intersection angle¶

From the examples shown in Table 4 to Table 7, we can see that the lifetime of inter-satellite communication for the different group of orbit planes are from couple of hundred seconds to about 18 hours. This fact will impact the routing technologies used for satellite network and will be discussed in Section 7.3.3.¶

7.3.3. Problems for Traditional Routing Technologies

When the satellite network is integrated with Internet by traditional routing technologies, following provisioning and configuration (see Figure 15) will apply:¶

1. The ground stations connected to local network and internet are treated as PE router for satellite network (called PE_GS1 and PE_GS2 in the following context), and all satellites are treated as P router.¶
2. All satellites in the network and ground stations are configured to run IGP.¶
3. The eBGP is configured between PE_GS and its peered network's PE or CE.¶

The work on PE_GS1 are:¶

• The local network routes are received at PE_GS1 from CE by eBGP. The routes are redistributed to IGP and then IGP flood them to all satellites. (Other more efficient methods, such as iBGP or BGP reflectors are hard to be used, since the satellite is moving and there is no easy way to configure a full meshed iBGP session for all satellites, or configure one satellite as BGP reflector in satellite network.)¶
• The internet routes are redistributed from IGP to eBGP running on PE_GS1, and eBGP will advertise them to CE.¶

The work on PE_GS2 are:¶

• The Internet routes are received at PE_GS2 from PE by eBGP. The routes are redistributed to IGP and then IGP flood them to all satellites. (Similar as in PE_GS1, Other more efficient methods, such as iBGP or BGP reflector cannot be used.)¶
• The local network routes are redistributed from IGP to eBGP running on PE_GS2, and eBGP will advertise them to Internet.¶

  /--------\             S1---S2----S3----\               /------\
 /          \            /    IGP domain   \             /        \
+ Local net CE--eBGP--PE_GS1---S4---S5---PE_GS2--eBGP--PE Internet +
 \          /            \                 /             \        /
  \--------/              \---S6---S7---S8/               \------/

Local access Internet through inter-satellite-networking¶

On PE-GS1, due to the fact that IGP link between PE_GS1 and satellite is not steady; this will lead to following routing activity:¶

1. When one satellite is connecting with PE_GS1, the satellite and PE_GS1 form a IGP adjacency. IGP starts to exchange the link state update.¶
2. The local network routes received by eBGP in PE_GS1 from CE are redistributed to IGP, and IGP starts to flood link state update to all satellites.¶
3. Meanwhile, the Internet routes learnt from IGP in PE_GS1 will be redistributed to eBGP. eBGP starts to advertise to CE.¶
4. Every satellite will update its routing table (RIB) and forwarding table (FIB) after IGP finishes the SPF algorithm.¶
5. When the satellite is disconnecting with PE-GS1, the IGP adjacency between satellite and PE_GS1 is gone. IGP starts to exchange the link state update.¶
6. The routes of local network and satellite network that were redistributed to IGP in step 2 will be withdrawn, and IGP starts to flood link state update to all satellites.¶
7. Meanwhile, the Internet routes previously redistributed to eBGP in step 3 will also be withdrawn. eBGP starts to advertise route withdraw to CE.¶
8. Every satellite will update its routing table (RIB) and forwarding table (FIB) after the SPF algorithm.¶

Similarly on PE_GS2, due to the fact that IGP link between PE_GS2 and satellite is not steady; this will lead to following routing activity:¶

1. When one satellite is connecting with PE_GS2, the satellite and PE_GS2 form a IGP adjacency. IGP starts to exchange the link state update.¶
2. The Internet routes previously received by eBGP in PE_GS2 from PE are redistributed to IGP, IGP starts to flood the new link state update to all satellites.¶
3. Meanwhile, the routes of local network and satellite network learnt from IGP in PE_GS2 will be redistributed to eBGP. eBGP starts to advertise to Internet peer PE.¶
4. Every satellite will update its routing table (RIB) and forwarding table (FIB) after IGP finishes the SPF algorithm.¶
5. When the satellite is disconnecting with PE-GS2, the IGP adjacency between satellite and PE_GS2 is gone. IGP starts to exchange the link state update.¶
6. The internet routes previously redistributed to IGP in step 2 will be withdrawn, and IGP starts to flood link state update to all satellites¶
7. Meanwhile, the routes of local network and satellite network previously redistributed to eBGP in step 3 will also be withdrawn. eBGP starts to advertise route withdraw to PE.¶
8. Every satellite will update its routing table (RIB) and forwarding table (FIB) after the SPF algorithm.¶

For the analysis of detailed events above, the estimated time interval between event 1 and 5 for PE_GS1 and PE_GS2 can use the analysis in Section 5.1. For example, it is about 398s for LEO and 103s for VLEO. Within this time interval, the satellite network including all satellites and two ground stations must finish the works from 1 to 4 for PE_GS1 and PE_GS2. The normal internet IPv6 and IPv4 BGP routes size are about 850k v4 routes + 100K v6 routes [BGP-Table-Size]. There are couple critical problems associated with the events:¶

P1. Frequent IGP update for its link cost

Even for satellites in different orbit with the steady relative positions, the distance between satellites is keep changing. If the distance is used as the link cost, it means the IGP has to update the link cost frequently. This will make IGP keep running and update its routing table.¶

P2. Frequent IGP flooding for the internet routes

Whenever the IGP adjacency changes (step 1 and 5 for PE_GS2), it will trigger the massive IGP flooding for the link state update for massive internet routes learnt from eBGP. This will result in the IGP re-convergency, RIB and FIP update.¶

P3. Frequent BGP advertisement for the internet routes

Whenever the IGP adjacency changes (step 3 and 7 for PE_GS1), it will trigger the massive BGP advertisement for the internet routes learnt from IGP. This will result in the BGP re-convergency, RIB and FIB update. BGP convergency time is longer than IGP. The document [BGP-Converge-Time1] has shown that the BGP convergence time varies from 50sec to couple of hundred seconds. The analysis [BGP-Converge-Time2] indicated that per entry update takes about 150us, and it takes o(75s) for 500k routes, or o(150s) for 1M routes.¶

P4. More frequent IGP flooding and BGP update in whole satellite network

To provide the global coverage, a satellite constellation will have many ground stations deployed. For example, StarLink has applied for the license for up to one million ground stations [StarLink-Ground-Station-Fcc], in which, more than 50 gateway ground stations (equivalent to the PE_GS2) have been registered [SpaceX-Ground-Station-Fcc] and deployed in U.S. [StarLink-GW-GS-map]. It is expected that the gateway ground station will grow quickly to couple of thousands [Tech-Comparison-LEOs]. This means almost each satellite in the satellite network would have a ground station connected. , Due to the fact that all satellites are moving, many IGP adjacency changes may occur in a shorter period of time described in Section 5.1 and result in the problem P1 and P2 constantly occur.¶

P5. Service is not steady

Due to the problems P1 to P3, the service provider of satellite constellation is hard to provide a steady service for broadband service by using inter-satellite network and traditional routing technologies.¶

As a summary, the traditional routing technology is problematic for large scale inter-satellite networking for Internet. Enhancements on traditional technologies, or new technologies are expected to solve the specific issues associated with satellite networking.¶