Your PC loses its internet connection because something in the chain between the browser and the remote server is resetting, and the most common culprits are Wi-Fi power management silently switching off your adapter, expired DHCP leases, buggy or outdated network drivers, and weak signal causing retransmission timeouts. Each cause produces a distinct failure signature, which means you can identify the real problem instead of reinstalling drivers at random. This explainer traces the full path a packet takes, quantifies how often each layer fails, and gives you threshold-based tests to isolate the fault.

The reset chain: what actually breaks a connection

A connection is not one thing. It is a stack of leases and states, each with an expiry timer. Your Wi-Fi adapter holds an association with the access point. Your PC holds a DHCP lease for its IP address. The router holds a NAT translation table entry mapping your internal address to a public one. The remote server holds a TCP socket. Any of these expiring or being cleared drops your connection, and the disconnect you see is just the topmost symptom.

Power management is the most underdiagnosed layer. Windows lets the network adapter driver power down the radio to save energy, a feature controlled per-device in Device Manager. When the adapter sleeps, the access point still believes the client is associated, so it keeps the association alive while your PC stops answering. The mismatch resolves only when traffic forces a wake, which is why the first page load after idle often fails and a refresh works.

Driver-level resets behave differently. A faulty driver can trigger a media disconnect event, where Windows tears down and rebuilds the adapter without warning. In Event Viewer under System, these appear as Event ID 27 from the source e1dexpress, rt640x64, or similar vendor names. A cluster of Event ID 27 entries timestamped to your disconnects is close to a diagnosis.

At the IP layer, DHCP leases default to 24 hours on most home routers but can be as short as 2 hours on some ISP-supplied gateways. At the halfway point, the client requests renewal. If the router is busy, overloaded, or has a full lease pool, renewal fails and the client drops to a 169.254.x.x address, which cannot route to the internet. At the transport layer, TCP retransmits unacknowledged segments on a backoff schedule starting near 1 second and doubling. After roughly 5 failed retransmits, the socket resets, which surfaces to you as a dropped download or a stalled video call.

How often each layer fails, in numbers

Wi-Fi links drop packets even when healthy. On a clean 5 GHz channel at -50 dBm signal, packet loss sits under 0.1 percent. At -70 dBm, a level many homes accept as "working," loss climbs into the 1 to 5 percent range, and TCP interprets that loss as congestion, halving its window and stalling throughput (RFC 5681). Below -75 dBm, association itself becomes unstable and roaming between access points causes multi-second outages.

Two-point-four gigahertz is worse for a different reason. The band is 83.5 MHz wide in the US and shared with microwave ovens, baby monitors, and every neighbor's router. A channel overlap of even 5 MHz between your network and a strong neighbor network raises retry rates measurably. The non-overlapping channels in the US are 1, 6, and 11; anything else overlaps by design.

Ethernet fails less but fails differently. A marginal cable with a damaged pair negotiates down from 1 Gbps to 100 Mbps, and a cable with a split pair can cause link flapping. NIC energy-efficient Ethernet, enabled by default on many Intel and Realtek gigabit controllers, can drop the link during idle periods on some switches, producing disconnects roughly every few minutes of inactivity.

Router failures are frequently thermal. Consumer routers in enclosed spaces routinely run their CPUs above 80 degrees Celsius under load. Above roughly 85 degrees, some SoCs throttle or reset the wireless radio. A router that drops connections only in the afternoon, or only after two hours of streaming, fits this pattern.

Edge cases that mimic disconnection

Some failures look like disconnects but are not. Captive portal re-authentication on hotel or campus networks drops your traffic at the gateway, not the link. A VPN client with a kill switch enabled will block all traffic the moment its tunnel drops, which feels identical to a dead connection but is the VPN protecting you. DNS failures are the most common false positive: if your resolver is unreachable, pages fail to load while ping to a numeric IP still works, and users report "no internet" when the link is fine.

Virtual adapters add another layer. Hyper-V, VMware, and VirtualBox install virtual switches that can capture or filter traffic. A misconfigured virtual switch can black-hole traffic for the host, and VPN clients install their own virtual adapters with their own power-management settings. Bluetooth interference is real but narrow: 2.4 GHz Bluetooth and Wi-Fi share spectrum, and a Bluetooth headset streaming audio near a 2.4 GHz-only laptop can raise retries noticeably on channels 1 through 6.

A counterintuitive detail professionals rely on: a connection can be "up" at the link layer while being unusable at the IP layer. Windows shows the network icon with a globe or warning precisely because the Network Location Awareness service probes for internet reachability. If that probe fails but the link is up, the problem is upstream of the adapter, which means adapter troubleshooting wastes time.

Misconceptions that waste hours

The biggest misconception is that disconnects equal a bad ISP. In a typical home, the ISP link is the most stable segment, and the local Wi-Fi link is the least. Restarting the router "fixes" things because it clears the NAT table and resets radios, not because the ISP link was broken.

A second misconception is that a new router solves everything. If your PC's adapter is set to aggressively power-save, or its driver is outdated, a new router changes nothing. The same applies to speed tests: a speed test measures throughput over a few seconds and cannot detect a disconnect that happens after 10 minutes of idle.

A third is that 5 GHz always beats 2.4 GHz. At long range through two walls, 5 GHz signal falls below -75 dBm while 2.4 GHz remains at -65 dBm, and the "slower" band is the stable one. Band steering on modern routers can also push devices onto a weaker 5 GHz band against their interest.

Diagnostic timeline: what you should see, and when

Time since symptom starts What you observe Likely layer Confirming test
0 to 2 seconds Page stalls, spinner, no error yet Wi-Fi retransmission Ping gateway, watch for loss
2 to 10 seconds Browser shows "no internet" DHCP renewal failure ipconfig /all, check lease
10 to 30 seconds Adapter shows disconnected Driver media disconnect Event Viewer, Event ID 27
30 seconds to 2 minutes Reconnects, works, drops again Power management or roaming Disable adapter power saving
2 to 10 minutes Repeated cycles under load Router thermal or overload Touch router case, check uptime
24 hours after setup Drops at a fixed time DHCP lease expiry Compare drop clock time to lease

This timeline reconciles the thresholds you will meet in guides. Event ID 27 within seconds means driver; lease renewal failure at a consistent interval means DHCP; drops that track idle periods mean power management. Each threshold points to a different fix, so test in order rather than applying all fixes at once.

Practical actions with thresholds

Start with a wired test. Connect Ethernet directly to the router and run a continuous ping to 1.1.1.1 for 10 minutes. Under 0.5 percent loss on wired means the fault is in your Wi-Fi path. Above 1 percent loss on wired means the fault is the router, cable, or ISP.

Check adapter power management. In Device Manager under Network adapters, open your adapter's Properties, Power Management tab, and clear "Allow the computer to turn off this device to save power." Then in advanced settings, set the wireless adapter's power-saving mode to maximum performance.

Check the DHCP lease. Run ipconfig /all and note the lease obtained and expiry times. If drops recur near expiry, set a static IP outside the router's DHCP pool, or shorten the lease in the router settings.

Check your signal and channel. Anything below -70 dBm on 5 GHz or -67 dBm on 2.4 GHz is unstable for sustained use. Move to a non-overlapping channel (1, 6, or 11 on 2.4 GHz) and disable band steering temporarily to force a stable band.

Update or roll back the driver. Use the vendor's site rather than Windows Update. If disconnects started after an update, roll back via Device Manager, Driver tab.

Red-flag warnings to take seriously: Repeated disconnect cycles with Event ID 27 after every driver update point to a hardware fault, not software. Router case hot to the touch and drops only under load: stop streaming tests, check ventilation. Adapter disappears entirely from Device Manager: power off, reseat the card, check for a failing slot. Drops only on one device while others stay connected: the fault is the device, not the network.

A comparison of the main remedies:

Remedy Fixes Misses Effort
Disable adapter power saving Idle disconnects Driver resets, DHCP 2 minutes
Update/roll back driver Media disconnect events Router-side issues 10 minutes
Static IP DHCP renewal failures Wi-Fi signal problems 5 minutes
Wired connection All Wi-Fi path faults PC-side driver bugs 1 cable
Router firmware update Router resets, thermal bugs Client adapter issues 15 minutes
Replace router Overload, thermal, aging hardware Anything client-side USD 60 to 200

Limits of self-diagnosis

No amount of local testing proves the ISP link is healthy. Your router's logs are the closest evidence: look for WAN link down events, PPPoE reconnects, or excessive retransmits. If the WAN interface drops, the problem is upstream and only your ISP can fix it. Likewise, interference is invisible to software and requires a spectrum analyzer or a site survey app to confirm.

This is general educational information about network troubleshooting, not a guarantee of repair or a substitute for professional service. If you rent equipment, contact your ISP before replacing it, since tampering can affect support terms. Direct unresolved hardware faults to a qualified technician.